Progress 09/01/23 to 08/31/24
Outputs
Target Audience:Research in AIFARMS has been disseminated to various target audiences; we have published several papers in peer-reviewed journals and presented at professional meetings attended by other academic scientists and industry partners. Through their engagement in AIFARMS research, undergraduate and graduate students and postdoctoral researchers receive training. The AIFARMS team has also reached farmers, extension agents, and industry through demonstrations and field days. We have targeted the public as an audience for our activities through appearances on local TV, local radio, national farmer conference/workshops and the Farm Progress show. AIFARMS, in conjunction with the 4 other National AI for Ag. Institutes hosted the AI for Agriculture research summit in Washington D.C. on July 29th and 30th, 2024. As part of this conference, participants from the AgTech sector, NGOs, researchers in AI and Agriculture, and government agencies were in attendance. AIFARMS has been featured on several national media outlets, broadening our target audience to include members of the wider public outside of agriculture. We are collaborating with several companies interested in our research to demonstrate and test practical applications and adapt our approaches to meet their needs. Changes/Problems:During Year 4 of AIFARMS, Dr. Supratik Guha, requested a change for the University of Chicago subaward. Dr. Guha assumed a new role at the university and had to shift time away from AIFARMS. Dr. Roser Matamla, who holds a joint appointment with Argonne National Laboratory and the University of Chicago, was appointed as the new sub-award lead for the University of Chicago. The statement of work she submitted during the change is included below. Statement of Work, The University of Chicago The Soil Thrust explores the potential of AI in utilizing soil data to make local and regional yield predictions. The work to be undertaken encompasses several tasks, including the final construction and deployment of the wireless underground sensor network (WUSN) for monitoring soil moisture, temperature, and electrical conductivity. It also involves sensor maintenance and the measurement of nitrogen, data analysis related to crop yields and soil properties, and nutrient analysis. It also entails developing and applying AI techniques for data gap-filling and correlation yield analysis. Construction and deployment of the Wireless Underground Sensor Network: The first task involves constructing and deploying the WUSN to monitor soil properties, such as soil moisture, temperature, and electrical conductivity. This network employs a LORA antenna and communication system. Currently, the motes and antennas are undergoing ground testing and will be ready for field deployment this Fall. The deployment team, consisting of four technicians, will position approximately fifty sensors and three antennas below the plowing depth. These motes will collect measurements every 30 minutes, resulting in a total of 48 readings every day. This translates to about 7,200 data points daily, which will serve as a valuable training dataset for numerous new AI techniques and applications. In particular, machine learning methods will be employed for data augmentation and gap filling, as well as enhancing local yield predictions by utilizing real-time soil properties in farming operations. Over the next two years, the budgeted technical support will facilitate the achievement of this goal, encompassing the deployment and maintenance of the WUSN. Maintenance of nitrogen sensors: This task involves maintaining nitrogen sensors and monitoring nitrogen levels in the field. Nitrogen sensors have been deployed in the field to collect data on run-off nitrogen using an automated nitrogen probe system. This system measures the amount of nitrate released from the field due to fertilizer applications, which then enters the subsurface water through the tiling system. Maintenance tasks include calibration, drift determination and cleaning of the nitrogen probes. Budgeted technical support will ensure the nitrogen probes remain in optimal working condition, allowing for consistent data collection. This data will contribute to our overarching goal of exploring the AI 's potential to use soil data for local and regional yield predictions while also considering nutrient releases within a field. Data analysis of crop yields and soil properties: The project aims to analyze the substantial volume of data generated by the WUSN and the nitrogen probes through the recruitment of two research fellows. One fellow will have expertise in soil physics, while the other will specialize in AI, ML techniques. Development and application of AI techniques: With the assistance of the research fellows and collaborations with AI experts Drs. Jingrui He and Hanghang Tong, a database and data training platform, will be established to develop and train new AI algorithms. Specifically, the focus will be on creating machine learning techniques for data augmentation and gap-filling applications, addressing common data gaps encountered with field sensors. Additionally, efforts will be made to develop algorithms that can enhance local yield predictions by incorporating soil real-time soil properties into farming operations. What opportunities for training and professional development has the project provided?Tenure-track faculty and senior researchers: AIFARMS has 40 senior researchers and faculty spanning universities and disciplines. Monthly seminars: Information is disseminated via monthly seminars. The whole AIFARMS team meets bi-annually for a team retreat and annual conference to discuss research progress, new directions, and accomplishments. Individual thrust meetings are held weekly or biweekly for each thrust, with sub-projects meeting on a similar schedule. Graduate and postdoc training: 49 students and postdoctoral scientists work on AIFARMS projects across 6 thrusts and 26 sub-projects. Trainees span CS and domain departments, with significant interactions between disciplines throughout regular meetings, special events, and social events throughout the year. In Year 4, 3 graduate students spent a semester in PhenoRob as part of the AIFARMS international collaboration. Undergraduate student training: AIFARMS sponsors summer REU students in conjunction with the Center for Digital Agriculture. In 2024, 22 students joined the joint program with 8 AIFARMS faculty mentors involved in mentoring the students. Thrust 1: Vikram Adve, Katie Driggs-Campbell, Girish Chowdhary, Yuxiong Wang (an integrated project with Thrust 3). Thrust 2: Angela Green-Miller ran the full program in collaboration with Christina Tucker and graduate students from the Green-Miller lab. Green-Miller and Isabella Condotta supported students. Thrust 3: Lisa Ainsworth Thrust 5: Sunoj Shajahan In addition to the REU program, several undergraduate students are involved in AIFARMS research projects throughout the regular school year. K-12 education: In Year 4, the education and outreach thrust has participated in multiple K-12 events. Over 200 K-12 students in Illinois were exposed to digital agriculture. Dr. Tucker and members of Tuskegee University hosted several teacher training events (including the one described below) with K-12 teachers to share the digital ag. in a box curriculum, providing hands-on experiences. Digital Agriculture K-12 Educator Workshop: Lessons in Creating a Digital Savvy Workforce: The Digital Ag in a Box workshop was successfully executed with K-12 educators in Alabama, attracting over 20 participants. Each educator received a Bit Rover Board, enabling them to engage in a hands-on activity to construct a miniature version of an agricultural robot. Several educators integrated the Bit Rover Board into their STEM curriculum. Feedback from the participants was overwhelmingly positive, demonstrating a keen enthusiasm to introduce Digital Agriculture concepts to their K-12 students. Develop an Introduction to Computer Science course for CS Teachers - K-12 educators in-service educators participated in an Exploring Computer Science workshop (9-12th) or CS Makers (K-8th) with AL teachers at TU. During these workshops, educators were trained on pedagogy deemed successful in teaching Computer Science in the classroom. 4-H: The goal of our digital agriculture programming has been to identify the equipment, technology, and resource capacity of Illinois youth involved with the 4-H programs. These activities can be applied to build college and career skills in the Digital Ag field. Our approach has been to collaborate with schools and 4-H staff to implement learning opportunities for middle and high school youth, focusing on hands-on learning experiences. Mark Becker leads a project at the Chicago High School for Agricultural Sciences. The knowledge sharing involved has focused on training CHSAS staff and students.The FarmBots from this project are now being incorporated in several service projects ongoing at CHSAS. Workforce Development: Many of the projects described above are significant investments in training the next generation of agriculture and AI professionals. In addition to those programs, AIFARMS hosted a summer AI foundry. The program's goal is to increase competency in tackling agriculture challenges with AI during a week-long hands-on course designed for graduate students with limited experience in machine learning. Following the 4-day training, students participate in a Hackathon sponsored by local AgTech companies. The program is split between AI topics (ML and CV), covered in the morning lectures, and agriculture applications (livestock management, crop improvement, and robotics) covered in the afternoon. 81 participants were involved in the 2024 AI foundry, including 5 faculty members of our 1890s capacity-building award partner - Langston University. In Spring 2024, Rachel Switzky (Siebel Center for Design) developed and ran the first iteration of the Design Thinking for Digital Agriculture course. The course enriches the university community by providing a unique interdisciplinary learning experience. Students from various fields, including engineering, agriculture, and business, gain valuable human-centered design and digital technologies skills, enhancing their competitiveness in the job market. Faculty benefit by integrating design thinking principles into their teaching and research, promoting a culture of innovation. This course will be shared with the broader community outside of UIUC, offering significant insight into agriculture through innovative solutions for real-world challenges that benefit farmers and agribusinesses. How have the results been disseminated to communities of interest?Engagement with the Research Community: The researchers on the team have at least 40 publications and participated in at least 32 conferences or presentations. Throughout Year 4, the AIFARMS team had the opportunity to showcase the impact and accomplishments of the institute through multiple public events and essential meetings. These events enabled the team to make connections with a broader community, initiating knowledge transfer of AIFARMS outcomes and promoting the federal investment in AI for Agriculture research. A few key events are described below. July 29-30, 2024, AIFARMS jointly hosted the AI for Agriculture Summit with the 5 USDA-NIFA national AI for Ag. institutes. The summit was a critical first step towards building a community of AI and Agriculture researchers, industry, and NGOs focused on addressing grand challenges food, fuel, and fiber challenges facing U.S. and world agriculture. This conference was funded by USDA-NIFA and AIVO. AIFARMS and the Environmental Defense Fundhosted a half-day workshop to discuss upcoming Opportunities and Challenges in Climate Smart Agriculture. NSF asked AIFARMS to participate in the AI Expo. Dr. Adve and Dr. Wedow presented at the AI Expo in D.C. to emphasize the role of foundational AI research in agriculture outcomes. In January 2024, Dr. Wedow presented for the Farm Foundation on the mission and accomplishments of AIFARMS. Throughout Year 4, AIFARMS was honored to host several high-profile visitors from the government and fellow academic institutions. In September of 2023, AIFARMS hosted a delegation from the PhenoRob Cluster of Excellence; we spent a week visiting with researchers in their labs and fields, exploring common research directions. In the spring, we were honored to be visited by a member of the UIUC Board of Trustees - Tami Craig Schilling (Bayer), to highlight AIFARMS and CDA research. In March 2024, AIFARMS, as part of the Center for Digital Agriculture, participated in the Center's annual conference with many members of the region's AgTech industry and academics from UIUC and neighboring universities present. Engagement with Government: September 2023, Drs. Adve and Bolden-Tiller participated in NSF Hill Day. All 25 National AI institutes were involved and had the opportunity to speak with congressional staffers during the event. In October of 2023, Tuskegee University hosted its annual Professional Ag. Workers Conference. Dr. Misra (director of USDA-NIFA) was in attendance and spent the afternoon with members of the TU and UIUC AIFARMS team. In February 2024, Dr. Wedow attended the USDA 100th annual Agricultural Outlook Forum. While there, she spoke with Josh Stull (USDA-NIFA) and Davida Tengey (USDA International Engagement Officer) regarding the AI for Ag. Institutes. This meeting led to the 5 institutes meeting with Dr. Tengey to discuss our international engagements Multiple NSF and USDA program officers participated in the AI for Agriculture Summit. Dr. Dionne Toombs, deputy director of USDA-NIFA, gave the keynote presentation for the summit. AIFARMS, in conjunction with all USDA AI Institutes, wrote a joint briefing paper for distribution to the general public and government agencies. The 1-page paper presents high-level benefits to support AI-Ag research. The current version can be seen in the appendix. Engagement with Industry: A collaborative project between the livestock thrust and Cargill explored the nutrition and health status of growing pigs with computer vision techniques. On-farm data (video) with Cargill on a commercial production farm with a focus on the behavior and temperament impact of different nutritional supplements. They extended the detailed behavior ethograms from the research setting to the commercial farm and developed a unique summary technique to represent the activity index. The CropWizard project led by Dr. Adve is currently in active discussion with numerous companies. We have highlighted the most promising connections below. Frenzy is interested in using CropWizard for their agronomists to find answers to technical issues in their work. Precision Planting, a division of AGCO, would use CropWizard with input from information gathered by their field technicians. Corteva is interested in licensing CropWizard and leveraging several of its components to enhance their Generative AI chatbot. Bayer, the CropWizard team, is currently discussing the scope of work with a shared license agreement. Additional companies the team is in conversation with are Cedar Bluff Technologies and John Deere. International Engagement: Partnership with PhenoRob Cluster of Excellence - DigiCrop - In the fall of 2023, AIFARMS in conjunction with CDA, PhenoRob Cluster of Excellence, Wageningen University, and ETH Zurich launched the joint DigiCrop network. The network currently has open exchanges of ideas, a common website, and plans to host DIGICROP conference in 2025. Three students (two form the autonomous farming thrust and one from the environmental resilience thrust) spent the summer in Germany working with members of PhenoRob and The University of Bonn. Engagement with MSIs: In Year 4, Langston University (AIFARMS sub-award) was awarded the Capacity Building Grant for "Cultivating Agricultural Leaders: Establishing Foundations for AI-driven Innovation in Sustainable Dairy Farming and Student Training." Dr. Angela Green-Miller and Dr. Condotta are Co-PIs on the award with Dr. Carlos Alvarado (Langston). This collaboration aims to expand capacity toward technological advancements using AI in dairy goat production systems. The AIFARMS faculty will provide consultations on the design, training, and implementation of the dairy goat monitoring systems at Langston University. This will include Langston students participating in the CDA-REU program. In addition to the CBG already funded with Langston, we have submitted letters of support for (1) South Carolina State University (Dr. Joe Mari Maja), (2) Lincoln University (Dr. Xukai Zhang), and (3) Prairie View A&M University (Dr. Ripendra Awal), and (4) 3 independent proposals from Tuskegee University (Dr. Perry, Dr. Chen, Dr. Bernard). Data management: The data management team, through collaboration with the Data Engineering for AI Applications working group that gathers data engineers and data managers across different AI institutes, established that each institute uses a different cyberinfrastructure to store and host its data. Although research artifacts are accessible, there is no mechanism through which the datasets published by different institutes can be easily found and retrieved. After having explored different ways through which the accessibility and findability of the artifacts that originate from the AI Institutes can be enhanced, the group has settled on testing the ML Commons Croissant format (mlcommons.org) for expressing the datasets. The adoption of a common format through which to express datasets has the potential to increase the findability, accessibility, and interoperability of the research artifacts that are published by different AI Institutes. Members of the working group are currently working on individually testing the format for their respective datasets and will be recommending the ways in which this format can be expanded and enhanced to accommodate the description of agricultural datasets. The AIFARMS data management group continued their work on providing access to datasets that were created as part of AIFARMS through a webportal (https://data.aifarms.org). We have worked on reformatting the PigLife dataset created and published by Thrust 2 in the ML Commons Croissant format and, therefore, making it more analysis-ready. Over the summer, we worked with an undergraduate student from the University of Michigan to improve the web portal. What do you plan to do during the next reporting period to accomplish the goals?In this section, we focus on the broad plans for Year 5, spanning the overall project. The team continues strengthening the synergy between foundational AI research and domain science. Throughout Year 4, the leadership team continued to encourage cross-thrust engagement and collaborations and cross-institute collaboration. Most of the subprojects within AIFARMS will continue into Year 5 as they have made satisfactory progress towards their stated goals. Their plans are described Overall Research section of the main report (which can be provided upon request). A few modified directions in the AIFARMS effort will be addressed within the next year. Dr. Nico Martin will be formally joining the AIFARMS teams. He has been a member of the Center for Digital Agriculture and has been involved in corresponding research projects. Dr. Martin will supervise PhD student John Meyers, a former member of the TU extensions department. Mr. Meyer's project will join work between UIUC and Tuskegee. His work will complement the 1890s sabbatical Dr. Bernard is completing at UIUC. This project will be a bridge between the work occurring on farms in AL and IL, with improved crop diagnostic tools using drone imagery. To merge outcomes within the environmental resilience thrust, Dr. Hudson, Dr. Lipka, and Dr. Ainsworth will be utilizing new techniques developed in the Hudson and Lipka labs to identify new genotype-to-phenotype relationship within a soybean Nested Association Mapping (NAM) population gathered at SoyFACE research center. This project will not only merge the work occurring between 3 AIFARMS researchers but the broader photosynthetic phenotyping research occurring at the USDA ARS SoyFACE Research Site. The soil monitoring and health thrust will investigate interesting and unexpected results from the soil microbiome work. This work will be led by Dr. Wendy Yang and Dr. Kiayu Guan. With advanced model-data fusion approaches, project researchers will study the relationship between soil nitrous oxide (N2O) and climate-smart management practices. The outreach and education thrust will continue to develop programs to excite and grow a new generation of students. In Year 5, the education team will partner with the Jackie Joyner-Kersee Foundation (jjkfan.org) to pilot the digital ag. in a box program with middle and early high school students. This program directly aligns with the goals of JJK Fan to show the connection between agriculture and physical/mental well-being. We will continue strengthening our international collaborations with the DigiCrop network (digicrop.net; CDA, PhenoRob Cluster of Excellence, Wageningen University, and ETH Zurich). We will jointly host DigiCrop 2025 under the DigiCrop network in July 2025. Over the past 4 years, we have developed a close network between the five National AI for Agriculture Institutes. We will continue to invest in this relationship. Specifically, we will be focusing on a joint Digital Agriculture workshop hosted with AgAID (WSU - lead) planned for 2025, hosted in Pune, India. This coordination is part of an EAGER proposal submitted by AgAID and partners (US-India-Japan) to evaluate India-centric agricultural applications. As we enter the last year of AIFARMS, we will focus on wrapping up research projects and determining the lasting outcomes of the AIFARMS Institute. We have drafted a visioning document for AIFARMS 2.0. We will continue to explore the proposed research agendas and new avenues throughout Year 5. The broad research objectives we are currently and will be exploring in the next year are outlined below. As part of this exploration, each thrust will hold a visioning workshop to discuss the next 5 years of AI for Agriculture (the environmental resilience team already held one). We will also submit mid-scale and large proposals to support the work for AIFARMS 2.0. Proposed Research Directions in AI for Crop Improvement Training recalcitrant biological systems with digital twins Using AI to optimize training sets for accurate genomic prediction in extreme future environments Transitioning AI-enabled biology from proof- of - concept to widespread adoption Proposed Research Directions in AI for Autonomous Farm Management Systems Generative AI for quantitative agriculture decisions Novel autonomous machine forms for production agriculture AI for smart agricultural implements AI for agricultural machine system optimization Proposed Research Directions in AI for Computer Vision and Robotics in Agriculture Open-world visual AI AI for autonomous agricultural robots
Impacts
What was accomplished under these goals?
During Year 4 of AIFARMS, the collaborations between AI and agriculture researchers produced several significant accomplishments, building on the cross-disciplinary relationships that started in Year 1. Foundational AI advances, relevant to the domain research across all research thrusts, have begun to mature, advancing agriculture in numerous ways, e.g., progress in autonomous navigation for ground robots, segmentation for livestock monitoring, ML-enabled genome segmentation, AI-enabled microbiome analysis, computer vision-based stomata detection, and generative AI for farm decision making. The five USDA-funded National AI Institutes for Agriculture hosted the AI for Agriculture Summit at the National Academy of Science in Washington, D.C. This joint conference was led by Dr. Jessica Wedow, Executive Director of AIFARMS. Dr. Wedow who secured funding and led a joint committee of all 5 institutes in planning the conference program. The conference convened researchers from all 5 USDA Institutes, 3 NSF AI Institutes, leading AI and Ag. Researchers, AgTech industry leaders, nonprofit research foundations, NGOs, NSF and USDA representatives. Dr. Dionne Toombs, Assistant Director of USDA-NIFA, opened the summit. A group of researchers from the five institutes are writing a white paper outlining AI for Agriculture research's current state, challenges, and future. A few critical outcomes from Year 4 of AIFARMS are highlighted. Education and outreach accomplishments are included in the summary of training and professional development component of this report. Foundational AI Research: Throughout the projects within AIFARMS, we focus on advances in foundational AI applied to agriculture; we have highlighted a few advances from Year 4 below. Previous reports described foundational zero-shot video segmentation work in AIFARMS, which enables many agricultural objects (pigs, chickens, corn ears, etc.) to be detected and classified without requiring any application-specific labeling. The latest extension of the segmentation shows object-level reasoning in deep net-based memory as opposed to pixel-level reasoning. Cheng, et. al. CVPR 2024 (selected as a CVPR 2024 highlight paper). https://arxiv.org/pdf/2310.12982 Robots struggle with plant management tasks since plants are fully deformable, with extensive occlusion, and vision sensors only observe a small fraction of their shape. We have developed a prototype robot system to manipulate leaves to reveal partially hidden fruits. For example, this can improve growth monitoring by providing automatic fruit volume estimation in later stages of growth. Zhang, et al., (CoRL), 2023, https://arxiv.org/pdf/2307.03175. Generative AI techniques have numerous potential applications in future agricultural management. We show experimentally that large language models (LLMs), despite being trained solely on text data, are surprisingly strong encoders for purely visual tasks in the absence of language, and we explain this phenomenon theoretically. This can be achieved by employing a frozen transformer block from pre-trained LLMs as an encoder layer to process visual tokens directly. Our approach consistently enhances performance across diverse computer vision tasks, including 2D and 3D visual recognition, non-semantic tasks, and multi-modal tasks. Pang, et al., ICLR 2024, https://arxiv.org/abs/2310.1297. Autonomous Farming Systems: Members of Thrust 1 launched the CropWizard project and service (https://uiuc.chat/cropwizard-1.5/). CropWizard is a Generative AI system that combines state-of-the-art multimodal Large Language Models (LLMs) with over 400,000 technical source documents from Extension units around the US to provide knowledge-grounded answers to agricultural questions. CropWizard is being made free of cost to Extension advisors nationwide and to academic users. One of the applications for this service is to augment the work of Extension advisors who can save significant time and effort when investigating answers to agronomics questions from farmers. Labor Optimization for Livestock Management: In the first 4 years of AIFARMS, the domain scientists within thrust 2 have implemented a real-time tracking algorithm for individually housed young pigs in the Piglet Nutrition and Cognition Laboratory (PNCL Facility). The data pipeline includes onsite raw video processing, GPU capabilities, and connectivity to the main campus for data storage. Current projects within AIFRAMS are using data for nutrition studies as a metric for animal status and exploring housing and husbandry practices. Within this facility, a significant dataset of images from young pigs over time (3 - 21 days) is under development and will be added to the PigLife datasets. The labs are developing models to recognize and track individuals despite changing physical size/shape/appearance characteristics. Environmental Resilience: Recently published work from the Hudson group (Zhang et al. 2024) identified >600K high-confidence genomic structural variants (SVs) in the Sorghum Bioenergy Diversity Panel. When these SVs were incorporated into downstream analyses, many results were improved over traditional methods, including the ability to resolve Sorghum and the identification of new GWAS SV-trait and SNP-trait associations.Preliminary results suggest that these improvements would unlock previously unexplored drivers of phenotypic variation in plants. While many attempts in the literature have been made to detect stomata cells and coastal zones automatically, numerous challenges have been identified in providing a generalized model. We proposed an automated strategy for the classification of microscope images by species with minimal labeled training data and active feedback control. This work is seminal because it overcomes the tedious process of labeling plenty of leaf images by experts for developing deep learning-based models. Automating the detection of stomatal cells in a leaf would help improve the overall crop yield. This work has been published in two highly selective computer vision conferences and is under review in two biology journals. Soil Monitoring and Health: A collaboration Plant Biology and CS has shown the potential for nitrification is surprisingly greater in the subsoil (below 15 cm depth) than in the surface soil. Our data suggest that competition for ammonium by heterotrophic microbes (which use soil organic carbon as a carbon source) may be relieved as soil organic carbon availability declines with soil depth, thereby allowing autotrophic nitrifiers (which use carbon dioxide as a carbon source) to have higher rates of activity below 15 cm depth. This work advances knowledge of potential rates of nitrification changes through the vertical soil profile. Understanding the relationship is essential to tracking nitrogen losses via leaching and nitrous oxide emissions. Technology Adoption and Public Policy: Thrust 5 have led a combined field-satellite level data integration for tracking cover crop adoption from data on field, farm, and farmer attributes over 2011-2021. The dataset included samples from 3995 farmers and 21,950 field parcels in the US Midwest (Illinois, Indiana, and Iowa) to analyze the determinants of cover cropping decisions at various scales. The results of this project aim to understand farmers' preferences for covering crop planting technologies. In the past year, members of our technology adoption thrust have been working closely with horseradish growers of southern Illinois to develop tools for non-chemical weeding. The project is an important collaboration between extensions, research, and growers, all integrating their knowledge to develop the technology needed for a small community. The robotic platform offers the flexibility to integrate mechanical implements for precise weed detection and removal in horseradish farms, with the transfer to other specialty crops planned as a future research direction.
Publications
- Type:
Conference Papers and Presentations
Status:
Under Review
Year Published:
2024
Citation:
I. Allabadi, G., Lucic, A., Wang, Y.X, & Adve. V. (2024). Learning to Detect Novel Animals with SAM in the Wild. International Journal of Computer Vision, Special Issue on Computer Vision Approaches for Animal Tracking and Modeling (IJCV 2024).
- Type:
Journal Articles
Status:
Under Review
Year Published:
2024
Citation:
Allabadi, G., Lucic, A., Yang, T., Wang, Y.X., & Adve, V. (2024). Dynamic Expansion and Adaptation for Open-World Semi-Supervised Object Detection.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
Chang, P., Liu. S., Ji, T., Chakraborty, N, Hong, K, & Campbell, K.D (2023). A Data-Efficient Visual-Audio Representation with Intuitive Fine-tuning for Voice-Controlled Robots. Conference on Robot Learning
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
Gasparino, M. V., Sivakumar, A. N. & Chowdhary, G. (2024). "WayFASTER: a self-supervised traversability prediction for increased navigation awareness." 2024 IEEE International Conference on Robotics and Automation (ICRA).
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
Kamtikar. S., Koe. K., Marri. S., Walt. B., Uppalapati. N., Krishnan., G. & Chowdhary., G. (2023). Visual Servoing for Pose Control of a Hybrid Continuum Manipulator in an Unstructured Environment. 7th Conference on Robotic Learning (CoRL) Learning for Soft Robotics Workshop.
- Type:
Journal Articles
Status:
Accepted
Year Published:
2023
Citation:
Liu, S. Gupta S. Wang. S., (2023). Building Rearticulable Models for Arbitrary 3D Objects from 4D Point Clouds Computer Vision and Pattern Recognition (CVPR).
- Type:
Journal Articles
Status:
Under Review
Year Published:
2024
Citation:
Man, Y., Gui, L.Y., & Wang, Y.X. (2024). Situational Awareness Matters in 3D Vision Language Reasoning. In processes. CVPR, 2024.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
Peng,.J.C., Yao, S., & Hauser., K. (2024). 3D Force and Contact Estimation for a Soft-Bubble Visuotactile Sensor Using FEM. IEEE International Conference on Robotics and Automation (ICRA), 5666-5672.
- Type:
Journal Articles
Status:
Accepted
Year Published:
2024
Citation:
Schreiber, A., Sivakumar, A. N., Du, P., Gasparino, M. V., Chowdhary, G., & Driggs-Campbell, K. (2024). W-RIZZ: A Weakly-Supervised Framework for Relative Traversability Estimation in mobile Robotics. IEEE Robotics and Automation Letters, 9(6), 56235630. https://doi.org/10.1109/lra.2024.3396095
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
Sivakumar, A., Gasparino, M., McGuire, M., Hisano Higuti, V., Akcal, M., & Chowdhary, G. (2024). Demonstrating CropFollow++: Robust Under-Canopy Navigation with Keypoints. [Paper presentation]. In Robotics: Science and Systems (RSS) 2024. (Nominated as finalist for Outstanding Demo Paper Award at RSS 2024 conference).
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
Sivakumar, A., Wang, N., G., Tommaselli, F., Gasparino, M., Becker, M., & Chowdhary, G. (2024). CropFollowRL: Learning Under-Canopy Navigation Policy with Keypoints Abstraction. In Workshop on Navigation & Mobile Manipulation in Challenging and Cluttered Natural Environments at Robotics Science and Systems (RSS) 2024.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
Sivakumar, A., Thangeda, P., Fang, Y., Gasparino, M., Cuaran, J., Ornik, M., & Chowdhary, G. (2024). Learning to Turn: Diffusion Imitation for Robust Row Turning in Under-Canopy Robots. Accepted as Extended Abstract to IEEE ICRA@40.
- Type:
Journal Articles
Status:
Accepted
Year Published:
2024
Citation:
Wang, Q., Downey, D., Ji, H., & Hope, T. (2024). Scimon: Scientific inspiration machines optimized for novelty. ACL 2024.
- Type:
Journal Articles
Status:
Accepted
Year Published:
2024
Citation:
Wang, X., Chen, Y., Yizhe, L., Zhang, Y. L., Peng. H., & Ji. H. (2024). Executable code actions elicit better llm agents. ICML 2024.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
Yao, S., Hauser, K. (2024). Structured Bayesian Meta-Learning for Data-Efficient Visual-Tactile Model Estimation. ICRA 2024 4th Workshop on Representing and Manipulating Deformable Objects, 2024, May 17).
- Type:
Journal Articles
Status:
Under Review
Year Published:
2024
Citation:
Ziqi, P., Xie, Z., Man, Y., & Wang, Y.X. Frozen Transformers in Language Models Are Effective Visual Encoder Layers. In-process. ICLR, 2024.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
Zhang, X., Gupta, S. (2023). Push Past Green: Learning to Look Behind Plant Foliage by Moving It. Conference on Robot Learning (CoRL).
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
Zhang, X. Chang, M. Kumar, P. Gupta. S. (2024). Diffusion Meets DAgger: Supercharging Eye-in-hand Imitation Learning. Robotics: Science and Systems (RSS).
- Type:
Journal Articles
Status:
Accepted
Year Published:
2022
Citation:
Li, J., Green-Miller, A. R., Hu, X., Lucic, A., Mahesh Mohan, M. R., Dilger, R. N., Condotta, I. C. F. S., Aldridge, B., Hart, J. M., & Ahuja, N. (2022). Barriers to computer vision applications in pig production facilities. Computers and Electronics in Agriculture, 200, Article 107227. https://doi.org/10.1016/j.compag.2022.107227
- Type:
Journal Articles
Status:
Accepted
Year Published:
2023
Citation:
Cao, S., Joshi, D., G, L.Y., & Wang, Y.X. (2023). HASSOD: Hierarchical Adaptive Self-Supervised Object Detection. Advanced in Neural Information Processing Systems, 36.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
Cao, S., Gu, J., Kuen, Tan, H., Zhang, R., Zhao H., Nenkova, A., Gui, L.Y., Sun, T., & Wang, Y.X, (2024). SOHES: Self-supervised Open-world Hierarchical Entity Segmentation. International Conference on Learning Representations (ICLR) 2024.
- Type:
Journal Articles
Status:
Accepted
Year Published:
2024
Citation:
Chen T., Wang, S., Guan, K., Jiang, C., Wu, X., Wang, S. (2024) Quantifying top-view crop residue cover from side-view imagery by geometric digital twins. (Under review).
- Type:
Journal Articles
Status:
Accepted
Year Published:
2024
Citation:
Gao L, Guan K, He L, Jiang C, Wu X, Lu X, Ainsworth EA (2024) Tropospheric ozone pollution increases the sensitivity of plant production to vapor pressure deficit across diverse ecosystems in the Northern Hemisphere. Science of the Total Environment 951: 175748.
- Type:
Journal Articles
Status:
Accepted
Year Published:
2024
Citation:
Hu, C., Zhou, Q., & Tong, H. (2024). Genius: Subteam Replacement with Clustering-based Graph Neural Networks. In S. Shekhar, V. Papalexakis, J. Gao, Z. Jiang, & M. Riondato (Eds.), Proceedings of the 2024 SIAM International Conference on Data Mining, SDM 2024 (pp. 10-18). (Proceedings of the 2024 SIAM International Conference on Data Mining, SDM 2024). Society for Industrial and Applied Mathematics Publications.
- Type:
Journal Articles
Status:
Accepted
Year Published:
2024
Citation:
Hyunsik Yoo, Zhichen Zeng, Jian Kang, Ruizhong Qiu, David Zhou, Zhining Liu, Fei Wang, Charlie Xu, Eunice Chan, and Hanghang Tong. (2024). Ensuring User-side Fairness in Dynamic Recommender Systems. In Proceedings of the ACM Web Conference 2024 (WWW '24). Association for Computing Machinery, New York, NY, USA, 36673678. https://doi.org/10.1145/3589334.3645536
- Type:
Journal Articles
Status:
Accepted
Year Published:
2024
Citation:
Jing, B., Yan, Y., Ding, K., Park, C., Zhu, Y., Liu, H., & Tong, H. (2024). Sterling: Synergistic Representation Learning on Bipartite Graphs. Proceedings of the AAAI Conference on Artificial Intelligence, 38(12), 12976-12984. https://doi.org/10.1609/aaai.v38i12.29195
- Type:
Journal Articles
Status:
Under Review
Year Published:
2024
Citation:
Tan, G.D., Chaudhuri, U., Varela, S., Ahuja, N, & Leakey, A. (2024). Machine learning-enabled computer vision for plant phenotyping: a primer on AI and case study on stomatal patterning. Journal of Experimental Botany. (Under Review)
- Type:
Journal Articles
Status:
Under Review
Year Published:
2024
Citation:
Varela, S., Zheng, X., Njuguna, J., Sacks, E., Allen, D., Ruhter, J., & Leakey, A. (2024). Breaking the barrier of human-annotated training data for machine-learning-aided plant research using aerial imagery. Journal of Plant Physiology. (Under Review)
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
Wu J., Ainsworth, E.A., Leakey, A., Wang, H. & He, J. (2023). Graph-Structured Gaussian Processes for Transferable Graph Learning. NeurIPS 2023.
- Type:
Journal Articles
Status:
Under Review
Year Published:
2024
Citation:
Wu, X., Guan, K., Wang, S., (2024) A generic framework to detect tillage practices from space: a demonstration in the US Midwest. (Under review).
- Type:
Journal Articles
Status:
Accepted
Year Published:
2023
Citation:
Wu, J., & He, J. (2023). A Unified Framework for Adversarial Attacks on Multi-Source Domain Adaptation. IEEE Transactions on Knowledge and Data Engineering, 35(11), 11039-11050. https://doi.org/10.1109/TKDE.2022.3230825
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
Wu, J., He, J., & Tong. H. (2024). Distributional Network of Networks for Modeling Data Heterogeneity. KDD 2024
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
Zeng, Z., Du, B., Zhang, S., Xia, Y., Liu, Z., & Tong, H. (2024). Hierarchical Multi-Marginal Optimal Transport for Network Alignment. Proceedings of the AAAI Conference on Artificial Intelligence, 38(15), 16660-16668. https://doi.org/10.1609/aaai.v38i15.29605
- Type:
Journal Articles
Status:
Accepted
Year Published:
2024
Citation:
Zhou, Q., Guan, K., Wang, S., Hipple, J., & Chen, Z. (2024). From satellite-based phenological metrics to crop planting dates: Deriving field-level planting dates for corn and soybean in the U.S. Midwest. ISPRS Journal of Photogrammetry and Remote Sensing, 216, 259273. https://doi.org/10.1016/j.isprsjprs.2024.07.031
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
Bao, W., Wei, T., Wang, H., & He, J. (2024). Adaptive test-time personalization for federated learning. In Proceedings of the 37th International Conference on Neural Information Processing Systems (NIPS '23). Curran Associates Inc., Red Hook, NY, USA, Article 3403, 7788277914.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
Ban, Y., Qi, Y., Wei, T., Liu, L., & He, J., (2024). Meta Clustering of Neural Bandits. KDD 2024.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2022
Citation:
Feng, S., Jing, B., Zhu, Y., & Tong, H. (2022). Adversarial Graph Contrastive Learning with Information Regularization. Proceedings of the ACM Web Conference 2022. https://doi.org/10.1145/3485447.3512183
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
Hill, B., Liu, L., & Tong, H. (2024). Ginkgo-P: General Illustrations of Knowledge Graphs for Openness as a Platform. WSDM 24: Proceedings of the 17th ACM International Conference on Web Search and Data Mining, 1066--1069. https://doi.org/10.1145/3616855.3635701
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
Qi, Y., Ban, Y., Wei, T., Zou, J., Yao, H., & He, J., 2024. Meta-learning with neural bandit scheduler. In Proceedings of the 37th International Conference on Neural Information Processing Systems (NIPS '23). Curran Associates Inc., Red Hook, NY, USA, Article 2796, 6399464028.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2024
Citation:
Wei, T., Jin, B., Li, R., Zeng, H., Wang, Z., Sun, J., Yin, Q., Lu, H., Wang, S., He, J., & Tang, X. (2024, March 15). Towards unified Multi-Modal personalization: large Vision-Language models for generative recommendation and beyond.
|
Progress 09/01/22 to 08/31/23
Outputs
Target Audience:Research in AIFARMS has been disseminated to various target audiences, we have published several papers in peer reviewed journals and presented at professional meetings attended by other academic scientists and industry partners. Through their engagement in AIFARMS research, undergraduate, graduate students and postdoctoral researchers are receiving training. The AIFARMS team has also been reaching farmers, extension agents and industry through demonstrations and field days. We have targeted the public as an audience for our activities through appearances on local TV, local radio, national farmer conference/workshops and the Farm Progress show. AIFARMS has been featured on several national media outlets, broadening our target audience to include members of the wider public, outside of agriculture. We are collaborating with several companies with an interest in our research to demonstrate and test practical applications and adapt our approaches to meet their needs. Changes/Problems:In January of 2023, Dr. Todd Mockler passed away. Dr. Mockler was the Danforth subaward lead and director of research for AIFARMS. In the spring of 2023, Dr. Andrea Eveland was appointed the new Danforth subaward lead. USDA-NIFA program manager, Steve Thompson, was made aware of this change and a revised statement of work for Danforth was submitted. The submitted statement of work follows, reporting for the Danforth subaward is included in section 5.7. Danforth Statement of Work: An existing project will be leveraged that is focused on developing predictive models for phenotyping crop canopy traits, including AI-derived canopy traits, and linking genotype to phenotype using maize and sorghum diversity. This began as a collaboration between Andrea Eveland, Todd Mockler (DDPSC), Girish Chowdhary, Alex Lipka (UIUC), and Vasit Sagan (SLU) (interfacing NSF-PGRP, IN2, and TGI projects) and uses remote sensing, computer vision, and machine learning (ML) for crop phenotyping and integrating above- and below-canopy traits as well as genomics and quantitative genetics for identifying loci underlying valuable crop traits and genomic predictions. Replicated trials of maize and sorghum diversity panels will be field-phenotyped. Summer 2023 will focus largely on maize as we've made significant progress in evaluating limitations and needs to maximize results. We've also collected extensive ground truth data over multiple years at UIUC and at the DDPSC Field Research Site (FRS) for leaf angle, leaf width, tassel architecture traits, LAI, and chlorophyll content, as well as physiological measurements such as thermal and stomatal conductance, and have preliminary results that show correlation with analysis of ground rover and UAV collected data. Maize will be grown in replicated blocks at UIUC and at the DDPSC FRS. Ground truth data will be collected for various canopy traits. Ground rovers (EarthSEnseTM) equipped with various sensors (LiDar, RGB, thermal) will be deployed to collect below/within-canopy data at several points throughout the growing season. Computational personnel in the Eveland lab will work with a graduate student in the Chowdhary lab (Junzhe Wu) and EarthSEnse to extract trait data from the RGB images and LiDar point clouds. We have begun to develop robust trait extractors for leaf angle and intend to extend analyses to multidimensional canopy features. In collaboration with the TGI, LiDar mounted backpacks and UAVs equipped with various sensors (RGB, LiDar, hyperspectral, thermal) will be used to collect data from within and above the canopy, respectively. A graduate student in the Eveland lab co-advised by Vasit Sagan will work on extracting trait information from these data using ML-based approaches. The project teams (Eveland/TGI, Chowdhary, Lipka, and other collaborating AIFARMS associates) will collaborate on 1) The development and implementation of new ML-based trait extractors for above- and below-canopy plant architecture traits; 2) The integration of below-canopy and above-canopy trait analyses for a comprehensive analysis of crop plant architecture; 3) The Implementation of ML/AI analyses to quantify novel crop traits that may not be tied to traditional physiological traits in crops, e.g., canopy features derived from analysis of discrete slices from canopy-level point cloud data; and 4) The integration of the ML-derived trait data into multi-trait GWAS analyses. The manual phenotype measurements will be used as training data during the development of ML trait extractors. Integration of above- and below-canopy traits captured by the ground robots and UAVs, with physiological traits through multi-trait association analyses, including the dynamic trajectories of these traits over the growing season, will enable us to associate specific plant architectures with physiological outcomes and potentially elucidate the genetic architectures that underlie these traits. Eveland's NSF-PGRP project with Lipka uses multi-trait GWAS in conjunction with biological information from genomics data and tissue-specific gene regulatory networks to identify genetic loci that underlie canopy architecture. Here, we will leverage these existing high-resolution genomics datasets along with the phenomics data collected in this project in ML-based approaches to refine and test SNP sets that can ultimately be used for genomic predictions. We will also leverage publicly available expression data that exists for subsets of these maize populations to aid in our predictive models. We will also extend this to field trials of sorghum. As part of Eveland's DOE-BER project, the 'DRIP' sorghum diversity panel was developed, which includes ~300 lines that maximize diversity in water use efficiencies as well as genetic and geographic diversity. All lines have been fully sequenced. This panel has been grown and phenotyped on the DDPSC Bellwether platform with a controlled drought regime. In the spring of this year, it will be grown and phenotyped under the gantry scanner at UA-MAC in AZ over an entire growth season in controlled well-watered and water limiting conditions. Samples have been collected from each of these lines for expression data as well as metabolites, which will enable genotype to phenotype associations. A major advantage of using sorghum for genotype to phenotype associations is the genome-level resolution that is attainable with the large number of sequenced genomes and pan-genome resource. However, trait extractors for canopy traits in the field will be more challenging due to tillering, etc. The DRIP panel will be grown out in replicated trials at UIUC and DDPSC FRS for above- and below-canopy phenotyping. Extensive phenotype data as well as ML pipelines and tools from other DDPSC-led initiatives including the ARPA-e TERRA-REF and OPEN projects will also be leveraged here and applied to field-collected data and used to link genotype to phenotype. This project will also interface with CABBI initiatives through collaborations with M. Hudson and A. Leakey as we have extensive genomics and phenomics data available across sorghum diversity. What opportunities for training and professional development has the project provided?In addition to the items illustrated in overall research and Products / Other Products, we have highlighted the following opportunities provided by AIFARMS: Students and postdoctoral scientist: 92 students and postdoctoral scientists work on AIFARMS projects across 6 thrusts and 25 sub-projects. Most students and postdocs work on multidisciplinary projects across AI and Agriculture disciplines and between thrust. Tenure-track faculty and senior researchers: AIFARMS has 45 senior researchers and faculty, 34 are tenure-track faculty. Faculty span disciplines including crop science, soil science, plant biology, animal science, human-centered design, agriculture and biological engineering, industrial engineering, and computer science. Monthly seminars: Information is disseminated via monthly seminars. These seminars are a mix of internal AIFARMS presentations, leadership updates, and external research presentations. The whole AIFARMS team meets bi-annually for a team retreat and annual conference to discuss research progress, new directions, and accomplishments. Individual thrust meetings are held weekly or biweekly for each thrust, with sub-projects meeting on a similar schedule. The thrust meetings are attended by students and postdocs who gain valuable experience in the running of multi-faceted projects. Master of Engineering Degree and associated certificates for Digital Agriculture: In the fall of 2022, AIFARMS, in conjunction with the Center for Digital Agriculture, received approval for the Master of Engineering in Digital Agriculture degree. The first cohort of students was admitted in spring 2023. The program's online delivery aims to provide continued training for working professionals, international students, and anyone interested in digital agriculture. The Digital Agriculture non-transcriptable certificate program was also approved.A marketing campaign is underway to increase the number of students to expand the program in Fall 2023 and beyond. Outreach: Outreach and exposure of modern autonomous farming and plant research tools to underrepresented students and small-scale farmers is an essential mission of AIFARMS. Tuskegee University and UIUC have participated in outreach events for government agencies, small-scale farmers, and students throughout Year 3. During the annual Professional Agriculture Workers Conference (Nov. 2022), Tuskegee faculty (Dr. Bernard, Dr. Chen, and Dr. Quansah) and UIUC Extension Educator Dennis Bowman met with small-scale farmers to provide a demonstration of autonomous tools (Terra Sentia rovers and cover cropping bot) to small-scale farmers from throughout the United States. Dennis Bowman also conducted numerous farmer-focused outreach activities in Illinois, Indiana, Iowa and Alabama, this year. Partnerships with community colleges led to 8 field day demonstrations of autonomous farming technologies reaching over 225 farmers and agriculture students. AIFARMS was also present at 9 regional trade shows, reaching 475 agricultural professionals and farmers. Bowman closed out the reporting year with two major outreach efforts. Over 900 Extension professionals learned about AIFARMS projects at the National Association of County Agricultural Agents Annual Meeting and Professional Improvement Conference. AIFARMS demonstrations were prominently featured at the University of Illinois College of ACES tent at the 2023 Farm Progress Show, the largest outdoor agricultural trade show in North America. Recruiting and participation of diverse students: Developing and broadening participation in digital agriculture careers has been a significant focus and accomplishment of AIFARMS since its inception. In addition to the launch of the Master of Engineering degree in Digital Agriculture, and significant accomplishments outlined in section 2.2 and Section 5.6 and 5.7, a summary of the highlights are included here. REU and undergraduate internships: AIFARMS and CDA supported 7 summer REU students in 2023 all from underrepresented groups and all females. Students were mentored by faculty in thrust 1,2, and 3. In addition, a UIUC student was sent to TU to complete a concurrent summer REU program. All REU students and TU undergraduate students, participating in programs at their university, were enrolled in the summer AI foundry for machine learning workshop. The second cohort of the CS teacher endorsement training had 34 teachers enrolled. We were able to secure funding to cover 80% of the tuition cost for all 34 teachers enrolled, with 32/34 of the teachers supported by a DPI grant. Tuskegee participated in several high school programs, undergraduate training events - including technical demonstrations, and k-12 educator trainings. Refer to section 5.7 for the full TU update. How have the results been disseminated to communities of interest?Engagement with the Research Community: The researchers on the team have at least 41 publications and participated in at least 40 conferences or presentations. On December 13th and 14th 2022, the AI Institutes Virtual Organization hosted the inaugural Summit for AI Institutes Leadership (SAIL). The summit gave AIFARMS team members a chance to meet with NSF leadership, partner AI institutions, and hold a joint meeting with the leadership of the three other USDA-NIFA funded AI-for-Ag institutes. Thrust 1 team member, Kris Hauser, helped kick off the event with his talk on case studies of AI applied "in the wild" - Toward Open-World AI for Autonomy. Vikram Adve spoke on a panel regarding achieving institute sustainability. Olga Bolden-Tiller and Rachel Switzky both spoke about expanding education and broadening participation. The leadership team made many new and essential connections while meeting many partners we had previously only spoken to in a Zoom setting. Adve and Wedow, with fellow AIFARMS PIs, wrote an overview article of AIFARMS for the AAAI Magazine special issue on the AI Institutes. The manuscript highlights the motivation, foundational AI, use-inspired AI research, and education efforts, and highlights the accomplishments of AIFARMS for a general audience. A key task undertaken by AIFARMS leadership has been to recognize the importance of "closing the loop" from agricultural use cases to use-inspired AI research goals to foundational AI challenges that must be tackled to accomplish those goals. Our extensive experience in the past three years with exploring how AI advances can be used to tackle a wide range of agriculture challenges has led to a far better understanding of the use-inspired and foundational AI research that will be needed. A major effort undertaken by the AI faculty in AIFARMS is to draft a vision paper describing the opportunities for AI researchers created by agriculture use cases, building on this extensive experience. The paper discusses seven broad and foundational AI research areas, describing, for each one, the motivating agricultural challenges, the key AI capabilities that will be required to tackle those challenges, the limitations on the current state of the art for those AI capabilities, and the data sets and resources that will be required to enable progress towards addressing those limitations. Close collaboration between AI and agriculture researchers will be essential to achieve these advances, and the paper ends with detailed recommendations for how agriculture researchers and policy makers can enable AI researchers to make progress towards meeting those challenges. Engagement with Industry: The AIFARMS team has pursued three active and one potential industry engagements in Year 3: with EarthSense, Microsoft, Cargill, and a potential one with Bayer. The EarthSense and Cargill partnerships are described under Thrusts 1, 2 and 5. The Microsoft partnership is described within the Tuskegee University update. On November 16th, 2022, Vikram Adve, Girish Chowdhary, and Jessica Wedow traveled to Bayer Global Headquarters in Leverkusen, Germany. AIFARMS leadership presented an overview of the ongoing research within AIFARMS. Throughout the presentation, members of the Bayer team asked engaging questions and proposed multiple possible next steps for collaboration. The areas of potential partnership include hyper-localized decision-making, farmer tradeoffs, and machine learning. This partnership is in development and will be a focus of Year 4. In addition to visiting with the team from Bayer Global, Tami Craig Schilling, VP of Global Market Development Agronomic Impact at Bayer, visited with AIFARMS and CDA faculty. With a generous gift from Bayer, AIFARMS was able to offer an award for the top three Hackathon solutions developed during the AI Foundry ML Hackathon. Engagement with Government: In November 2022, members of the AIFARMS team, in conjunction with the Center for Digital Agriculture, hosted officials from the U.S. Government Accountability Office (GAO). The GAO officials are part of an independent, nonpartisan agency working to inform congress. The visit will inform the GAO of technologies on precision agriculture in response to a provision in the Chips and Science Act. The AIFARMS and CDA team present an overview of our work and high-level research talks on 1. precision digital technology for crops, 2. machine vision and automation, and 3. a panel discussion on the impact of adoption. These presentations and discussions will inform an assessment of current precision agriculture technologies, their benefits, barriers to adoption, and related challenges. Later in the year, Tuskegee faculty and staff (Dr. Bernard, Dr. Quansah, Mr. Myers - Extension) served as panelists during the GAO Engagement on Precision Agriculture Technologies, where they discussed the concerns and constraints affecting small-scale farmers. Data Management: In Year 3, the AIFARMS Data Management (DM) working group has put significant effort into dissemination of work to various communities of interest. As part of this effort, the DM group completed a Dataset and Software Creation Best Practices tutorial. The DM team has also continued to expand the data sets hosted on the AIFARMS data portal (data.aifarms.org/). Following the conversations started at the SAIL conference, the data management team started an AI Agriculture focus Data Management working group with members from five other AI Institutes, including representatives from AIIRA, AIFS, AgAID, ICICLE in addition to the AIFARMS Institute. The group started monthly meetings in March of 2023. The goals of this working group are to ensure that the datasets the AI Agriculture Institutes release are made available, findable, accessible, and interoperable. To achieve this goal, we plan to look at the metadata that is collected for the actual data and either identify an existing suitable standard or create a new standard for the metadata. The group will create an index data portal that would host links to the datasets that are hosted at different institutions or repositories. This working group is essential to continue learning from each other's experiences and practices. International Engagement: In the winter of Year 3, AIFARMS launched an international partnership with PhenoRob Cluster of Excellence, University of Bonn. In November 2022, a delegation from AIFARMS, including Tuskegee University and the University of Chicago, traveled to the University of Bonn to launch a new partnership with PhenoRob Cluster of Excellence. The two partners met for a week of detailed research discussions, facility tours, and sharing of research motivations to outline future collaborations. Preliminary discussions occurred for a joint research network centered around a common place to disseminate knowledge related to common areas of interest. Throughout the course of Year 3, Madhu Khanna and Shadi Atallah have met with the PhenoRob team on a regular basis to write a joint research paper for the Annual Review of Resource Economics. The paper titled "Economics of Adoption of Artificial Intelligence-Based Technologies for a Sustainable Agriculture" is being co-authored with Thomas Heckelei and Hugo Storm from PhenoRob. Khanna and Atallah are also collaborating on a perspective piece with Hugo Storm titled "Steering Digital Technology Adoption and Usage Towards a Safe and Just Operating Space" Kris Hauser, member of the autonomous farming thrust, has been partnering with members of the 'Humanoid Robots Lab' at Bonn for plant manipulation with robotic arms to perform active crop reconstruction and mapping. Within Year 3, multiple PIs and their graduate students discussed new collaborations with members of the PhenoRob team. This project will involve graduate student exchange, to make progress on new directions in Year 4 to advance the work. What do you plan to do during the next reporting period to accomplish the goals?In this section we focus on the broad plans for Year 4, spanning the overall project. The team continues to strengthen the synergy between foundational AI research and domain science. Throughout Year 3, the leadership team continued to encourage cross-thrust engagement and collaborations. Within Year 4, we will continue to facilitate cross-thrust collaborations including a new sub-project between thrust 1 and 5.New directions and some reorganization of research are outlined below. Generative AI tools based on Large Language Models (LLMs) and computer vision foundation models have advanced at a remarkable pace over just the past 3-4 years. Recognizing the potential of these models, AIFARMS has launched a new project, CropWizard, to develop an interactive query answering tool for agriculture professionals, based on generative AI tools. We believe that such a tool, trained on public multimodal data sources including the scientific literature, undergraduate and graduate educational materials, extension reports, and other sources of practical guidance, can benefit all the major communities of agriculture professionals and researchers. A key challenge we will tackle is to improve the reliability of the responses it generates by increasing the domain understanding of the core ML models underlying the tool. Precision agriculture solutions do not fit the needs of specialty crops within the US. Large equipment companies have a lack of interest or incentive to develop tools for specialty farms; however, AIFARMS is well positioned to address the essential needs of specialty farms. In the fall of 2023, AIFARMS will launch a new sub-project to tackle autonomous weeding in horseradish. The team will develop a reference architecture for AI based precision agriculture solutions for autonomous weeding in horseradish. This project will be a joint effort between thrust 1 and thrust 5. The education thrust will form a new sub-project geared toward middle school students to interest them in applying computer science and technology for climate-smart agriculture tools. The 'Digital Ag in a Box' will generate learning modules and hands-on activities. The modules will focus on building fluency in programming for agricultural robotics applications and outfitting demonstrations with sensors to monitor environmental variables. The AIFARMS team will complete the writing of the visioning paper on AI opportunities in agriculture. In 2024, AIFARMS in conjunction with the other four USDA AI-Institutes will host an AI-Agriculture visioning conference. Our goal is to develop a common vision across the USDA-NIFA funded AI Institutes for how the AI research community can advance solutions for agriculture and ag-related climate challenges. We will focus on the big-picture ideas and challenges in AI research relevant to agriculture, emphasizing but not exclusively focusing on research at our five institutes. The conference will feature high-profile speakers from the AI and Ag communities, funding agency representatives, and some government representatives. In Year 4 we will expand upon the creation of open-source datasets for use in foundational AI research. The advisory board was excited about the promise of this dataset to generate a unique open-access dataset for use as a 'benchmark' for model comparison. Members of thrust 2 will continue to expand the annotation within this dataset and work with members of thrust 1 to generate a useful tool for the AI community. In Year 4, the soils thrust will integrate three complete years of crop yields and soil data, and two additional years of complete C and N fluxes with the online soil, tile and GHG sensor. The dataset incorporates biogeochemistry domain knowledge to identify (i) cause of yield variance a sub-field scale, (ii) prediction C and N flux hotspots moments, and (iii) test hypothesized soil-crop-water causal linkages too complex for traditional, non-ML analyses. The soil monitoring and health thrust will additionally, explore new approaches to mine data using AI-based approached from the scientific literature, addressing the small data problem of soil biogeochemistry and providing a mutual advancement for domain and AI science. Spurred by the conversation around the PigLife dataset, members of thrust 3 and thrust 1 have begun to take steps to generate a set of datasets curated around well annotated phenotypic data generated for application in foundational AI challenges. This project will pull from existing partnerships at UIUC (DOE-CABBI, USDA-ARS-SoyFACE, DOE-Terra Ref) and expand to the phenotyping community throughout the US and world. In Year 4, the data management team, will create a dataset geared towards algorithmic work on the "open world" object detection problems that tackle the issue of the model being able to recognize previously unseen and unknown objects and incrementally learning them without forgetting the old classes of objects that the model was originally trained on. Datasets that are geared towards this problem are scarce. Under the auspices of Thrust 1, initial work has been conducted through which we have identified several individual plant and animal datasets that can be merged and adapted for algorithmic work that would tackle the open world object detection. An immediate use case that can benefit from such a dataset is precision livestock management monitoring that typically collects a large amount of unlabeled videos and images. Being able to recognize a previously unseen species in the video stream and to incorporate it into the ongoing precision management would improve the models that are geared towards tracking and monitoring livestock. The plan for Year 4 is to evaluate the TERRA REF dataset that is hosted at the NCSA with respect to making a subset of the dataset more accessible and available for algorithmic work and the plant phenotyping community. In the coming year, we will continue to explore the synergy between AIFARMS and the additional USDA- funded projects at Illinois (Farm of the Future and iCOVER). Part of this expansion will be a collaborative Industry Partnership program between AIFARMS and CDA. Dr. John Reid will be developing this program with support from NCSA industry and current AIFRMS partners. Additional broad goals for Year 4 have been identified by the leadership team: Continue active discussions with companies that are ongoing in order to identify areas of mutual interest and joint collaborative projects. Expand the support for our REU and K-12 educational programs. In Year 3, thrust 2 completed an on-farm research trial with Cargill that will be expanded in Year 4. The engagement with EarthSense (ES) will continue to grow in Years 4 and 5. We will work with ES to evaluate improvements in robot autonomy in commercial robots, as well as new capabilities for coordinated robot teams. The cover cropping work will also continue to scale up through the partnerships with the I-FARM and iCOVER projects. Continue to support the public release of AIFARMS datasets hosted on the AIFARMS data portal. Expand the REU program to support a larger cohort of students, recruit diverse domain and AI students, and expand the number of recruited universities. Develop and expand the connections between the other national AI institutes. Joint collaborations between the other four USDA funded institutes The data management team will continue to connect with Agriculture associated AI institutes. UIUC -INVITE. AIFARMS, MMLI, and INVITE will explore collaborations in Year 4. International network of prominent digital agriculture institutes. In the fall of 2023, AIFARMS in conjunction with CDA, PhenoRob Cluster of Excellence, Wageningen University, and ETH Zurich will launch the joint DigiCrop network.
Impacts
What was accomplished under these goals?
In the third year of operation, the AIFARMS Institute has continued building on the strong foundation laid in the program's first two years. Currently, AIFARMS supports 26 sub-projects, with an additional project added in year 3 within the environmental resilience thrust. AIFARMS will add two additional sub-projects in Year 4 to expand on the significant accomplishments made within the first three years. In January of 2023, a vital member of the AIFARMS team, Dr. Todd Mockler, passed away. Todd was an essential member of AIFARMS and served as a strong collaborator, associate director of research, and thrust 3 lead. Todd's positive impact on science will always be remembered. Foundational AI Integrated framework for segmentation applicable across agriculture-based challenges, including autonomous robotics, disease detection and livestock tracking. Mask2Former - Improved general architecture for many semantic segmentation tasks (Cheng et al. 2022). XMen - Video segmentation with limited memory and applicable to long videos (Cheng and Schwing 2022). Significant improvements in open-world object detection accuracy benefit agricultural use cases such as livestock detection and tracking, and visual detection for robotic harvesting. In collaboration with thrust 3, developed KEBLM (Knowledge-Enhanced Biomedical Language Models), a novel framework for incorporating various types of knowledge from multiple sources into biomedical PLMs. We studied a new problem named generalized few-shot node classification, where the test samples can be from both novel classes and base classes. We proposed a shot-aware graph neural network (STAGER) equipped with an uncertainty-based weight assigner module for adaptive propagation. This resulted in one paper published at ICDM 2022, which has been selected as a best-ranked paper (Zhang et al). Autonomous Farming Systems Significant effort and advances in handling occlusions in agriculture domains Improved video segmentation and tracking in the presence of occlusions Progress toward actively manipulating plants to minimize occlusions Improved robustness to occlusion within crop rows with sensor fusion techniques for operating in more robust field settings. Significant effort and advances in handling occlusions in agriculture domains New algorithms for video segmentation and multi object tracking in the presence of occlusions, including the re-appearance of objects (Choudhuri et al. 2023). Improved robustness to occlusion within crop rows with sensor fusion techniques for operating in more robust field settings. (Gasparino et al. 2023). Labor Optimization for Livestock Management Developed a comprehensive swine dataset, 'PigLife', to create an open-source benchmark dataset for developing and evaluating robust computer vision algorithms for complex pig production systems. This dataset aims to support the development of algorithms that focus on fundamental tasks, such as counting, detecting, tracking, and behavioral recognition of pigs. Developed and deployed deep learning models with high accuracy: computer vision models were developed to (a) segment in an image, (b) detect video view (top, front, side, other) (c) detect postures and (d) detect body regions in an image, especially its ears. Model output shown in figure 5. This will further analysis of temporal video data for correlation with reproductive status and identification of pigs. Thrust 2, jointly with thrust 6, was awarded an NSF REU Site: Drivers for Machine Learning and AI practices. The program will aim to mentor and foster research exploration for students in machine learning and cross-disciplinary teams within the biological sciences and engineering. Environmental Resilience Thrust members developed a cross-scale sensing technology to integrate ground measurements, airborne hyperspectral imagery, and satellite Earth Observation to scale and accurately quantify regional-scale information of management practices including cover crop, aboveground biomass, and tillage practices. This highly accurate and granular management information is important for field-level farming carbon intensity and sustainable agriculture assessment (Wang et al. 2023). Multivariate time series (MTS) imputation is an important problem for heterogenous data fusion. Existing methods fail to fully utilize the graph dynamics for precise imputation in more challenging MTS data such as networked time series (NTS). Extensive evaluations demonstrate the effectiveness of our method by outperforming powerful baselines for both MTS imputation and link prediction tasks on various real- world datasets. With in-depth analysis of synthetic trait-assisted Genomic Selection (GS), we determined that including synthetic traits with a biological trait of interest in a multi-trait GS model can increase the predictive ability of the biological trait of interest. Soil Monitoring and Health Finalization of the integration of three years of crop yields and soil data, from external funding, at the same site, while completing two additional years of complete C and N fluxes with the online soil, tile and GHG sensor to complete a biogeochemistry model. In Year 3, the team finalized wireless soil sensor design that had been delayed due to the COVID-19 pandemic and supply chain shortages. The sensors are currently in the process of deployment. We completed data collection for the microbiome structure and function sub-project, a major accomplishment given both the large number of samples and the large number of soil, microbial, and biogeochemical variables measured. Technology Adoption and Public Policy In the past year, thrust 5 completed a choice experiment survey of 450 farmers to examine incentives to adopt cover cropping and the role that the attributes of a cover crop planting robot can be expected to play in inducing greater adoption. Based on this work, midwestern corn farmers are more likely to adopt a cover cropping technology if it suppresses weeds and increases short-term (1 - 3 yrs) and long-term (>4 yrs) yields and will allocate more acreage to cover cropping if neighbors adopt the practice. (Zhang et al. 2023) Education and Outreach Continued advocacy and support for CS legislation to establish a competitive grant program to support the development and enhancement of CS programs in K-12 schools. AIFARMS in conjunction with CDA hosted a successful year 2 of the AI Foundry for Agriculture including tracks for livestock and crop-based ML applications, with the final Hackathon sponsored by Bayer. In the spring of 2023, AIFARMS and CDA launched the Master of Engineering in digital agriculture and associated certificates. Key Accomplishment for the Donald Danforth Plant Science Center The Eveland lab developed a trait extractor for automatically detecting and quantifying leaf angle from a field robotic system equipped with low-cost sensors. Key Accomplishment for Michigan State University Yield stability zones (YSZ) provide an effective and practical integrative measure of the small-scale variability of soil health. We tested this by determining the values for various metrics of soil health from fields in samples replicated across YSZ, using a soil test suite used by producers. We also investigated the covariation of field-scale management with the degree of heterogeneity of these soil health parameters. We found that the use of YSZ allowed us to successfully partition soil organic carbon and soil health into significantly separate areas, with the low and stable yield zones statistically lower in normalized SOC when compared to the high and stable yield zones and unstable zones. Key Accomplishments for the Tuskegee University Industry collaboration with Microsoft to develop Tuskegee University's Farm of the Future, FarmVibes sensors, to integrate satellite based biophysical products and field soil sensor data to help develop data and information-based precision agriculture.
Publications
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
Wu, L. (July 2023) The Annual Meeting of Applied and Agricultural Economics Association, Washington DC. "Drivers and barriers to adopting cover cropping technology among midwestern farmers.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
Zhao, Y., Sharif, H., Pao-Huang, P., Shah, V., Sivakumar, A. N., Gasparino, M. V., Mahmoud, A., Zhao, N., Adve, S., Chowdhary, G., Misailovic, S., Adve. V. (2023). ApproxCaliper: A Programmable Framework for Application-aware Neural Network
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
Zhou, Q., Wang, S., Liu, N., Townsend, P., Jiang, C., Peng, B., Verhoef, W., Guan, K.,(2023). High-performance atmospheric correction of airborne hyperspectral imaging spectroscopy: model intercomparison, key parameter analysis, and machine learning surrogates. ISPRS Journal of Photogrammetry and Remote Sensing. (Accepted)
- Type:
Journal Articles
Status:
Published
Year Published:
2023
Citation:
Li, N., Zhou, S., Margenot, A.J. 2023. From prairie to crop: spatiotemporal dynamics of surface soil organic carbon stocks over 167 years in Illinois, U.S.A. Science of the Total Environment. 159038.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
Fernandes, S.B. (2023) Improving Genomic Prediction with Synthetic Traits. Presentation at the annual NCCC-170 Research Advances in Agricultural Statistics meeting at the University of Wisconsin-Madison, June 15-16, 2023.
- Type:
Journal Articles
Status:
Under Review
Year Published:
2023
Citation:
Gasparino, M. V., A. E., Higuti, Sivakumar, A. N., Velasquez, A. E. B., Becker, M., & Chowdhary,
G. (2023). CropNav: A Framework for Autonomous Navigation in Real Farms. Under review at IEEE International Conference on Robotics and Automation.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
Walt. B., Krishnan, G. (2023) Grasp State Classification in Agricultural Manipulation. [submitted] IROS 2023
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
Wu, L. (April 2023) The program in Environmental and Resource Economics, ACE, UIUC. "Drivers and barriers to adopting cover cropping technology among midwestern farmers.
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Gasparino, M. V., Sivakumar, A. N., Liu, Y., Velasquez, A. E., Higuti, V. A., Rogers, J., ... & Chowdhary, G. (2022). Wayfast: Navigation with predictive traversability in the field. IEEE Robotics and Automation Letters, 7(4), 10651-10658. https://doi.org/10.1109/LRA.2022.3193464
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Ji, T., Dong, R., & Driggs-Campbell, K. (2022). Traversing Supervisor Problem: An Approximately Optimal Approach to Multi-Robot Assistance. Robotics: Science and Systems (RSS). https://doi.org/10.48550/arXiv.2205.01768.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2021
Citation:
Jing, B., Feng, S., Xiang, Y., Chen, X., Chen, Y., & Tong, H. (2021). X-GOAL: Multiplex Heterogeneous Graph Prototypical Contrastive Learning. Proceedings of the 31st ACM
International Conference on Information & Knowledge Management. (CIKM2022), 894-904. arXiv. https://doi.org/10.1145/3511808.3557490.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Jing, B., Yan, Y., Zhu, Y., & Tong, H. (2022). COIN: Co-Cluster Infomax for Bipartite Graphs.
NeurlPS 2022 GLFrontier Workshop. arXiv. https://doi.org/10.48550/arXiv.2206.00006
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Cheng, B., Misra, I., Schwing, A. G., Kirillov, A., & Girdhar, R. (2022). Masked-attention mask transformer for universal image segmentation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 1290-1299). https://doi.org/10.48550/arXiv.2112.01527
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Cheng, H. K., & Schwing, A. G. (2022, October). Xmem: Long-term video object segmentation with an atkinson-shiffrin memory model. In European Conference on Computer Vision (pp. 640-658). Cham: Springer Nature Switzerland. https://doi.org/10.48550/arXiv.2207.07115
- Type:
Journal Articles
Status:
Under Review
Year Published:
2022
Citation:
Cisneros-Velarde, P., Lyu, B., Koyejo, S., & Kolar, M. (2022). One Policy is Enough: Parallel Exploration with a Single Policy is Near-Optimal for Reward-Free Reinforcement Learning. arXiv. https://doi.org/10.48550/arXiv.2205.15891
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Jing, B., Zhang, S., Zhu, Y., Peng, B., Guan, K., Margenot, A., & Tong, H. (2022). Retrieval Based Time Series Forecasting. 31st ACM International Conference on Information & Knowledge Management. (CIKM2022), AMLTS Workshop. arXiv. https://doi.org/10.48550/arXiv.2209.13525
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2022
Citation:
Jing, B. Zhang, S., Zhu, Y., Peng, B. Guan, K., Margenot, A.J., Tong, H. 2022. Retrieval Based Time Series Forecasting. arXiv: 2209: 13525.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2022
Citation:
Kamboj, A., Ji, T., & Driggs-Campbell. K. (2022). Examining Audio Communication Mechanisms for Supervising Fleets of Agricultural Robots. IEEE International Conference on Robot & Human Interactive Communication (RO-MAN).
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Li, J., Green-Miller, A. R., Hu, X., Lucic, A., Mahesh Mohan, M., Dilger, R. N., Condotta, I. C., Aldridge, B., Hart, J. M., & Ahuja, N. (2022). Barriers to computer vision applications in pig production facilities. Computers and Electronics in Agriculture, 200, 107227. https://doi.org/10.1016/j.compag.2022.107227
- Type:
Journal Articles
Status:
Accepted
Year Published:
2023
Citation:
Li, N., Bullock, D., Butts-Wilmsmeyer, C, Gentry, L., Goodwin, G., Han, J., Kleczweski, N, Martin, N., Paulausky, P., Pistorius, P, Seiter, N., Schroeder, N., Margenot, A.J. (2023). Distinct soil health indicators are associated with on-farm variation in maize yield and tile drain nitrate losses across contrasting nitrogen applications in central Illinois. Soil Science Society of America Journal. Accepted.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Wang, Q., Li, M., Chan, H. P., Huang, L., Hockenmaier, J., Chowdhary, G., & Ji, H. (2022). Multimedia Generative Script Learning for Task Planning. arXiv. https://doi.org/10.48550/arXiv.2208.12306
(Submitted to ICLR 2023)
- Type:
Journal Articles
Status:
Published
Year Published:
2023
Citation:
Wang, S., Guan, K., Zhang, C., Jiang, C., Zhou, Q., Li, K., Qin, Z., Ainsworth, E. A., He, J., Wu, J., Schaefer, D., Gentry, L. E., Margenot, A. J., & Herzberger, L. (2023). Airborne hyperspectral imaging of cover crops through radiative transfer process-guided machine learning. Remote Sensing of Environment, 285, [113386]. https://doi.org/10.1016/j.rse.2022.113386
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2022
Citation:
Wang, H., Huang, W., Wu, Z., Tong, H., Margenot, A.J., He, J. 2022. Deep Active Learning by Leveraging Training Dynamics. NeurIPS. Accepted.
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Wang, S., Guan, K., Zhang, C., Lee, D., Margenot, A, J., Ge, Y., Peng, J., Zhou, W., Zhou, Q., and Huang, Y. 2022. Using soil library hyperspectral reflectance and machine learning to predict soil organic carbon: Assessing potential of airborne and spaceborne optical soil sensing. Remote Sensing of Environment. https://doi.org/10.1016/j.rse.2022.112914
- Type:
Conference Papers and Presentations
Status:
Awaiting Publication
Year Published:
2022
Citation:
Wasserman J., Yadav K., Chowdhary G., Gupta Abhinav, Jain Unnat, Last Mile Embodied Visual Navigation, Conference on Robot Learning (CORL), Auckland NZ, Dec. 2022
- Type:
Journal Articles
Status:
Published
Year Published:
2023
Citation:
Li, N., Zhou, S., Margenot, A.J. 2023. From prairie to crop: spatiotemporal dynamics of surface soil organic carbon stocks over 167 years in Illinois, U.S.A. Science of the Total Environment. 159038. https://doi.org/10.1016/j.scitotenv.2022.159038
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Wei, T., You, Y., Chen, T., Shen, Y., He, J., & Wang, Z. (2022). Augmentations in Hypergraph Contrastive Learning: Fabricated and Generative. NeurlPS 2022. arXiv. https://doi.org/10.48550/arXiv.2210.03801
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Wu, J., He, J., Wang, S., Guan, K., & Ainsworth, E.A. (2022). Distribution-Informed Neural Networks for Domain Adaptation Regression. NeurIPS 2022.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Wu, J., He, J., & Ainsworth, E. (2022). Non-IID Transfer Learning on Graphs. Accepted by Conference on Artificial Intelligence (AAAI-23). arXiv. https://doi.org/10.48550/arXiv.2212.08174
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2022
Citation:
Wu. J., He. J, Wang. S., Guan. K., Ainsworth E.A. (2022). Distribution-Informed Neural Networks for Domain Adaption Regression. Proceedings of the Advances in Neural Information and Processing Systems 35 (NeurIPS 2022).
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Wu, J., Tong, H., Ainsworth, E., & He, J. (2022). Adaptive Knowledge Transfer on Evolving Domains. In S. Tsumoto, Y. Ohsawa, L. Chen, D. Van den Poel, X. Hu, Y. Motomura, T. Takagi, L. Wu, Y. Xie, A. Abe, & V. Raghavan (Eds.), Proceedings - 2022 IEEE International Conference on Big Data, Big Data 2022 (pp. 1389-1394). (Proceedings - 2022 IEEE International Conference on Big Data, Big Data 2022). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/BigData55660.2022.10020944
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Yan, Y., Zhou, Q., Li, J., Abdelzaher, T., Tong. H. (2022). Dissecting Cross-Layer Dependency Inference on Multi-Layered Inter-Dependent Networks, Proceedings of the 31st ACM International Conference on Information & Knowledge Management. (CIKM2022), 2341-2351. https://doi.org/10.1145/3511808.3557291
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Zhang, Z., Du, B., & Tong, H. (2022). SUGER: A Subgraph-based Graph Convolutional Network Method for Bundle Recommendation. Proceedings of the 31st ACM International Conference on Information & Knowledge Management. (CIKM2022), 4712-4716. arXiv. https://doi.org/10.48550/arXiv.2205.11231.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2022
Citation:
Zhou, Y., Wu, J., Wang, H., & He, J. (2022). Adversarial Robustness through Bias Variance Decomposition: A New Perspective for Federated Learning. 31st ACM International Conference on Information & Knowledge Management. (CIKM2022), arXiv. https://doi.org/10.48550/arXiv.2009.09026
- Type:
Journal Articles
Status:
Accepted
Year Published:
2023
Citation:
Zhou, Q., Wang, S., Liu, N., Townsend, P., Jiang, C., Peng, B., Verhoef, W., Guan, K., High- performance atmospheric correction of airborne hyperspectral imaging spectroscopy: model intercomparison, key parameter analysis, and machine learning surrogates. ISPRS Journal of Photogrammetry and Remote Sensing. (Accepted).
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Allabadi, G., Lucic, A., Pao-Huang, P., Wang, Y. X., & Adve, V. (2023). Semi-Supervised Object Detection in the Open World. https://doi.org/10.48550/arXiv.2307.15710
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
Azam, R. N., Fernandes, S. B., El-Kebir, M., Koyejo, S., Lipka, A. E., Leakey, A. (2023). Towards Automating Highly Heritable Phenotype Discovery For Plant Breeding. Poster at Computational Approaches to Scientific Discovery Workshop 2023 AAAI Spring Symposium
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Bao, W., Wang, H., Wu, J., & He, J. (2023). Optimizing the Collaboration Structure in Cross-Silo Federated Learning. https://doi.org/10.48550/arXiv.2306.06508
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2021
Citation:
Choudhuri, A., Chowdhary, G., Schwing, A. G.. Proceddings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021. pp.13598-13607. Optimization, Sixth Conference on Machine Learning and Systems (MLSys). Miami, Fl, 2023.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Cisneros-Velarde, P., Lyu, B., Koyejo, S., Kolar, M. (2023). One Policy is Enough: Parallel Exploration with a Single Policy is Near-Optimal for Reward-Free Reinforcement Learning. [not formally published, but its acceptance] http://aistats.org/aistats2023/accepted.html
|
Progress 09/01/21 to 08/31/22
Outputs
Target Audience:Research in AIFARMS has been disseminated to various target audiences, we have published several papers in peer reviewed journals and made presentations at professional meetings attended by other academic scientists. Through their engagement in AIFARMS research, undergraduate and graduate students as well as postdoctoral fellows are receiving training. The AIFARMS team has been reaching farmers, Extension agents and industry through media interviews, podcasts, demonstrations and field days. We have also targeted the public as an audience for our activities through appearances on the local TV and the Farm Progress sShow. AIFARMS has been featured on several national media outlets, broadening our target audience to include members of the general public. We are collaborating with several companies with an interest in our research to demonstrate and test practical applications and adapt our approaches to meet their needs. Changes/Problems:
Nothing Reported
What opportunities for training and professional development has the project provided?In addition to the items illustrated in section Products / Other Products, we would like to highlight the following opportunities provided by AIFARMS: Students and post-docs: 45 students and post-doctoral scientists work on AIFARMS projects, spread out across all the AIFARMS research thrusts, with several of them working between agriculture and AI disciplines and between thrusts. Tenure-track faculty and senior researchers: Of the 43 senior researchers working with AIFARMS, 38 are tenure-track faculty or senior researchers, spread out across the six thrusts. There is strong interaction among the tenure track faculty from disparate areas--soil science, plant biology, animal science, human- centered design, and computer sciences as joint projects have begun. This has enabled a new degree of necessary interdisciplinarity to help researchers involved in agriculture going forward. Monthly seminars: Information is disseminated via monthly seminars across AIFARMS and individual thrust meetings which are held weekly or biweekly for each thrust. These thrust meetings are often attended by students and post-docs who gain valuable experience in the running of complicated, multi-faceted projects. Development of new Masters of Engineering Degree for Digital Agriculture: In Year 2 AIFARMS, in conjunction with The Center for Digital Agriculture, received approval for the Master of Engineering in Digital Agriculture degree, and the first cohort of students will be admitted for the spring 2023 semester. The program's online delivery aims to provide continued training for working professionals, international students, and anyone else interested in digital ag. The Digital Ag non-transcriptable certificate program has also been approved, and proposals for additional certificate themes are pending approval. Outreach: Outreach impacts training and professional development in both directions. This year there have been several outreach efforts with AIFARMS researchers participating in farmer focus groups, livestock farm visits, and related events. Thrust 5 conducted two farmer focus groups to determine the incentives, barriers and concerns related to adoption of mechanical weeding and cover cropping with and without autonomous robotic systems. These interactions have helped the training of farmers in techniques developed by AIFARMS researchers and helped AIFARMS researchers' professional development from insights gained from non-academic customer environments. Recruiting diverse students: The AIFARMS REU program, which has now run successfully for two years in a row, focused primarily on minority students from 1890 institutions, as described below. Moreover, several AIFARMS team members work closely with the National Society of Minorites in Agriculture, Natural Resources and Related Sciences (MANRRS). Dr. Bolden-Tiller has served as MANRRS president and is currently very active within the organization. Many of the outreach and development opportunities, including the FarmBot and CS teacher training activities, serve underrepresented minority populations, including those in Chicago and the Metro East St. Louis region of Illinois. iCAN program: iCAN was designed to addresses both the talent gap in tech and the underrepresentation of these groups in CS. The iCAN program provides foundational training in CS to agriculture students without said training at Illinois and partner institutions. The program is aimed at students interested in professional development or obtaining a graduate degree in the Digital Agriculture program, thus bringing more students with agriculture backgrounds into computing. REU and undergraduate internships: AIFARMS funded 12 summer REU interns in 2022 to work within research thrust 1, 2, 3, and 4. These students receive a stipend, as well as housing and meals for two months. Nine students are from historically black colleges and universities, which include Tuskegee University and Langston University. Additionally, four students are supported with scholarships from Microsoft and Corteva industry partners. AI Foundry for Agricultural Applications Summer Workshop: Dr. Angela Green-Miller and Dr. Isabella Condotta, members of thrust 2 (Livestock Management), hosted a summer workshop and Hackathon event. Students participated in lectures and virtual activities on topics focused on AI and ML in agriculture applications. Students were mentored by faculty from the Departments of Agricultural and Biological Engineering, Animal Sciences, and industry partners. The program aimed to teach skills relevant to many agricultural applications, and the datasets used in the course were specific to livestock systems. On the last two days of the course, participants were challenged to develop a solution to a digital agriculture problem in an inspiring Hackathon. How have the results been disseminated to communities of interest?Engagement with the Research Community: Results have been disseminated to academic communities of interest through conference presentations and publications in a broad range of disciplinary and interdisciplinary venues. For example, the researchers on the team have at least 30 publications and participated in at least 28 conferences and presentations. A number of these publications and presentations are in top-tier conferences and journals with a wide audience range, including high performance computing, programming systems, computer vision, digital agriculture, plant phenotyping, data science, robotics, agricultural economics and livestock welfare, amongst others. Broad student and faculty representation across annual conferences was seen despite university imposed COVID-19 travel restrictions. See the "Products" and "Other Products" section for a list of all publications and presentations. A few members of the AIFARMS team gave prestigious keynote presentations or plenary talks at several research-focused conferences, including at the World Soils Congress, North American Plant Phenotyping Network (NAPPN) Annual Conference, the Artificial Intelligence, Machine Learning and Data Science World Forum 2022, and the International Workshop on Machine Learning on Graphs at the Conference on Web Search and Data Mining (2022). Engagement with Industry: Members of the AIFARMS team have had active discussions and meetings with personnel from several companies, including EarthSense, AGCO, Microsoft, Honeywell, John Deere, IBM, Verizon, T-Mobile, Corteva, and Indigo Ag. Most of these conversations are at an early exploratory stage, except EarthSense, which is a thriving and close collaboration, and AGCO which is further along in discussions. EarthSense is closely engaged with the autonomous farming and technology adoption thrusts. Adoption of two technology outcomes from the AIFARMS research, has begun. One outcome is the advances in learning for control, which were developed in thrust 1 to enable robust autonomous navigation in outdoor settings using computer-vision instead of (more expensive and yet less reliable) LIDAR+GPS. A second broad outcome is being developed in direct collaboration with EarthSense, in particular, autonomous under-canopy robots for planting cover crops before harvest, which leads to improved yields and significantly lower planting costs for the cover crops. This area of work builds on research on improved autonomy in thrust 1 and on ongoing research on the economics of cover crop adoption in thrust 5. Less than 5% of all croplands in the United States uses cover crops, despite their well-documented and substantial benefits, and increasing this adoption even by a small extent would be a significant win. In addition to the technology transfer discussions with companies, both Corteva and Microsoft have provided scholarships for students in the summer REU program this year. These companies were invited to mentor the students during the summer and to attend the end-of-summer REU project presentations. We aim to expand the pool of company scholarships in future years. Engagement with Government, Media, and Policy Research Organizations: Dr. Andrew Leakey, member of thrust 3, testified before the House Science, Space and Technology Subcommittee of the U.S. Congress at a hearing on "Bioenergy Research and Development for the Fuels and Chemicals of Tomorrow." This testimony included highlighting the synergy resulting from partnerships between the Center for Advanced Bioenergy and Bioproducts Innovation (CABBI) and AIFARMS. We have had several high-profile media outlets publish articles regarding the work in AIFARMS, including The Wall Street Journal, Wired, and Market place report on National Public Radio. Dr. Vikram Adve, PI of AIFARMS, has participated in a number of policy-focused events and meetings, including meeting with the Illinois state director of agriculture and his team; the Computing Community Consortium (CCC) Climate Roundtable's discussion group on digital agriculture; the Microsoft research faculty summit workshop on digital agriculture and the Future of Food; a panel discussion organized by the communications of the ACM on AI for Science; a AAAI Panel on the AI Institutes; a workshop on AI / Data Science for collaboration between land grant universities organized by the US National Academies of Sciences, Engineering and Medicine (NASEM) board on agriculture and natural resources; and a presentation to the FCC connectivity needs working group on future demands for rural broadband. Dr. Bruno Basso co-led a workshop titled, "Reducing the Health Impacts of the Nitrogen Problem: An Environmental Health Matters Workshop, Digital Agriculture to Reduce Nitrogen Losses across the U.S. Corn Belt," organized by the US National Academy of Sciences, Engineering and Medicine (NASEM). Dr. Jessica Wedow made a presentation on AIFARMS to the IoT4Ag Center - Seed of Innovation Seminar Series on July 27th, 2022. This seminar series brought together research centers focused on AI, digital, and precision agriculture to spark collaborations between the AI institutes and partnering academic institutions. We also reached the agricultural community more directly through participation at the Farm Progress show, interviews with the local national public radio station, the University of Illinois Agronomy Day demonstration and through specific focus groups with growers. Data Sharing and Management: The Data Management working group has created a set of slides for best practices in dataset creation accessible to the whole AIFARMS team. These slides are meant to be a quick reference point for guiding thrusts throughout the dataset building process starting with data collection stages, continuing with different levels of documentation, selection of the dataset and software license, to preservation and publication of the dataset. The AIFARMS Data Portal was launched in August 2022. The creation of this portal serves as a secure storage option for the AIFARMS team, allowing for long term storage, computational services, and between thrust and institution data sharing. The goal of this portal is to make a unified data storage portal for the team to increase collaboration at all levels. We believe this portal will be beneficial to multi-institutional projects that would benefit from shared computational and storage platforms. The Data Management team is working with individual researchers and thrusts to format, store and preserve the data that are in various stages of the preparation in a safe and secure manner, along with making many of them that are ready to be shared with a wider community openly available. Currently AIFARMS has published 4 open source datasets, with many more currently in the process of publication. For more examples, please reference "Products" and "Other Products". What do you plan to do during the next reporting period to accomplish the goals?AIFARMS has made significant progress on our goals in Year 2. We will continue to strive for more synergy between algorithmic, computational and practical agricultural technology within our thrusts. Continued cross-project collaborations were encouraged by our first in-person meeting since AIFARMS was formed, in March of 2022. We held break-out and group sections to discuss collaborations between subprojects and what institute leadership can do to foster these connections. Year 3 will expand upon these cross-thrust connections. Most subprojects will continue as planned into Year 3. Of note are some broad new directions, outlined below. A major new direction we will begin in Year 3 is an international collaboration with the PhenoRob Cluster of Excellence in Germany, funded by a supplemental award of $300,000 from USDA-NIFA. The goal of this effort is to lay the foundation for long-term collaborative research efforts between the PhenoRob Center and AIFARMS in areas of common interest. PhenoRob core projects are developing new technologies complementary to those in AIFARMS, and collaboration can bring solutions of significant value to US agricultural. Examples include 3D object reconstruction for robotic harvesting, chemical-free mechanical weeding, improved predictions from hyperspectral data, technology adoption studies complementary to those used in AIFARMS, and techniques for analyzing root performance. We will use the funding supplement to organize activities required for developing longer-term collaborations, including planning workshops; exchange visits; sharing of complementary data sets; proposal development; and joint publications. We will explore synergy between AIFARMS and three other large, projects at Illinois, newly funded by the USDA over the past 2 years: COALESCE (jointly funded by NSF and USDA-NIFA), DIRECT4AG and the Farm of the Future. A broad new effort we are continuing is to review our foundational AI research challenges in the light of the experience and initial results of the first two years of AI-for-agriculture research activities. For these seven foundational AI challenges, we are discussing how these challenges arise in specific AI-for-Agriculture problems, to develop a more detailed understanding of the research priorities we must tackle to make progress on the collaborative AI+agriculture research. A key goal of these discussions is to ensure that our research budget allocations are appropriately aligned to ensure progress on these research challenge areas. Rachel Switzky, Director of the Siebel Center for Design at UIUC, will be joining the team. In thrust 6, Switzky will expand the scope of the MEng degree in digital agriculture by adding a foundation in 'design thinking'. Her team will design a new course and review existing courses for the MEng degree. In thrust 1, Switzky will work with the members of the Human-centered Autonomy subproject to tackle key questions centered on how humans interact with advanced technologies, working with the natural language communication and knowledge applications for agricultural robots and the FarmBot. Thrust 3 will slightly reorganize their efforts to focus on three coordinated using a sorghum population. Over the last two years, AIFARMS Danforth site lead, Dr. Todd Mockler, has sequenced the genomes of over a third of a sorghum diversity panel. This panel is being phenotyped in Dr. Andrew Leakey's group at UIUC. These related data sets will enable the subprojects below to coordinate their efforts closely. We have shifted brought in two new AI faculty members, Dr. Yuxiong Wang and Dr. Mohammed El-Kebir, both from the UIUC Department of Computer Science, to enhance the focus of these projects. Dr. Wang is a specialist in open-world AI problems while Dr. El-Kebir is a specialist in AI techniques for genomics and bioinformatics. Sensor-to-phenotype predictions: We will continue what was done in Year 2 and bring in Dr. Yuxiong Wang to work with Dr. Andrew Leakey and Dr. Narendra Ahuja on this subproject. (Dr. Wang is also collaborating on thrust 1, where we are studying open-world computer vision tasks, such as object detection and segmentation.) Genome sequencing: This area has shifted directions and brought in the additional support of Dr. May Berenbaum, Department of Entomology and Plant Biology. The focus of this work will be on improved AI-driven techniques for genome sequence completion. Informed GWAS: This subproject has reorganized to focus on AI for GWAS, informed by the phenotype information of subproject 2 and the genome sequences of subproject 4, i.e., the outcomes of the above biological focus areas. This is an expansion of current efforts within the thrust with the addition of Dr. Mohammed El-Kebir. Dr. David Bullock, UIUC Department of Crop Science, will begin work on a new subproject with Dr. Hanghang Tong, an expert in heterogeneous data mining. This project will center around the Illinois Extension - Data-Intensive Farm Management project (DIFM) project, led by Bullock, to solve challenges with heterogeneous data sets. The DIFM project is structured as a loop, involving systems of sensors, real-time and just-in-time computational analysis, model-based information use, and upscaling made possible by technology that generates, processes, analyzes, ad acts upon digital information. Other general goals for Year 3 include the following: Continue active discussions with the companies that are ongoing in order to identify areas of mutual interest for joint collaborative projects. Previously, the AIFARMS team received sponsorships from two companies to support REU students we will expand this effort in Year 3. AIFARMS personal have meet with senior personnel at AGCO, during Year 3 we will continue these discussions and focus on collaborative areas. In Year 2, joint evaluation of autonomous planting of cover crops with Earthsense lead to 100 acres planted. The effort will grow to over 1000 acres in Year 3. Additional joint focus groups for farmers and development of an economic survey through a partnership with thrust 5. Several other preliminary discussions are under way with John Deere, ADM, EarthSense, and PepsiCo, as well as a few production livestock farms. Develop new connections with other national AI institutes, especially the other three funded by the USDA. Support public releases of AIFARMS data sets: The Data Management Working Group will continue to develop an AIFARMS data portal, which began last year, designed for both internal and external data sharing. They will work to facilitate making data sets and software tools available publicly through the data portal. Ramping up the capacity and usage of our experimental facilities, especially the Illinois Autonomous Farm and the new Illinois Farm of the Future: Much of the robotics, edge computing, and sensing research takes place or is tested on these facilities. The growth of these facilities in Year 3 will support an expanded range of synergistic projects, all of which will contribute to strengthening the outcomes and impact from AIFARMS. Explore collaborations between two institutes centered at UIUC, the NSF-funded Molecular Maker Laboratory Institute and DOE- funded Center for Advanced Bioenergy and Bioproducts Innovation. All three institutes share faculty and have complementary research goals. Currently, talks between AI researchers from both AIFARMS and MMLI around the broad topic of generative modeling are scheduled for early in Year 3. Additionally, CABBI and AIFARMS have plans to share research datasets interesting to both institutes and collaborate on several data management objectives. Expand the REU program to be able to support a larger cohort of students, including obtaining outside funding to reach more students, specifically from MSIs.
Impacts
What was accomplished under these goals?
In Year 2 of operation, the AIFARMS Institute has continued to build on the strong foundation laid in Year 1 for a broad and impactful 5-year research program. The subprojects within the research thrusts bring together AI and Agriculture researchers in close collaborative efforts to explore how foundational advances in AI can impact important challenges in Agriculture. The list of subprojects has grown, with additional subprojects within the Soil Health and the Education and Outreach thrusts. The Education and Outreach thrust is contributing to important, impactful efforts for inspiring the younger generation to explore digital agriculture and to help grow a skilled digital agriculture workforce. A successful in-person REU program was hosted this summer, giving students hands-on research experience and career mentoring. Building on last year's successes with external research funding, several team members led a winning proposal to bring the USDA-NIFA Farm of the Future to the UIUC campus. Moreover, UIUC's role in hosting the AIFARMS Institute and our close partnership with Tuskegee University were both cited as key reasons for the win in an intensely competitive field. This project (I-FARM - Illinois Farming and Regenerative Management) will provide a valuable near-production-scale testbed to evaluate and demonstrate research outcomes from other major research efforts, including AIFARMS. Multiple new industry engagements launched this year or are currently under discussion with the potential to translate technical innovations from the group's research into commercial products. Our partnership with EarthSense is furthest along, with significant technology transfer successes in the areas of improving robot autonomy and cover crop planting. Team members have been invited to speak at many prominent research and education events and organized several community events. More specifically, some of the major accomplishments of the AIFARMS Institute in the second year are as follows. (Only the most important are included, due to space limitations; other accomplishments are listed in the complete report.) Details of the activities and outcomes that led to these accomplishments are described in Sections 2 and 3. Foundational AI Research We originally identified "foundational AI research challenges", which are common across many different domains, not specific to agriculture. In the second half of Year 2, we began a series of discussion meetings on these six research challenges (plus a 7th, Computer Vision). We also broadened the scope of the "Small Data Problem" to the more general class of "Open World AI". Advances on Open World AI (formerly called the Small Data Problem) Investigated the fundamental difference between convolutional neural networks that use complex-valued and real-valued parameters. There exists specifiable input to output mappings that complex-valued networks can perform but real-valued networks cannot. A novel transfer learning method based on Neural Network Gaussian Process to translate models from simulated to real-world data. Novel ML methods called heterophilic graph contrastive learning, which outperform previous supervised and SSL methods. Motivated by predicting soil N and P fluxes from soil datasets but with limited labeled data. Foundational advances in Computer Vision Novel framework for hierarchical (human) activity recognition from videos using deep neural networks New techniques for Heterogeneous Data Fusion (Also Foundational advances in Computer Vision) Significant progress on audio-visual input segmentation, by combining audio and vision data sources for improved segmentation results. Advances in Federated Edge Learning on Resource-constrained Devices Novel application-aware approximation tuning techniques to greatly reduce computational and energy requirements of neural networks while preserving end-to-end application quality goals. This was used to optimize vision-guided robot navigation and is now being explored for object manipulation (e.g., harvesting). Advances in Learning for Control CropFollow algorithm to enable vision-guided navigation without LiDAR or GPS for Level 2 autonomy. Strategies for Integrating Domain Knowledge into ML Efficient literature mining technique for entity linking in scientific literature is better than state-of-the-art methods and uses 60x fewer parameters. New techniques for Human Interaction with Autonomous Systems A new approach, Multimedia Generative Script Learning, to generate the next steps based on the goal and previous steps with visual scenes depicting the states of objects. Autonomous Farming Systems: This year, we demonstrated level 2 autonomy in field robots using both LIDAR + GPS and alternatively using (much cheaper and more robust) computer vision-guided models for navigation within crop rows. With the achievements in robot autonomy and foundational AI research, over 100 acres of cover-crops were planted well before harvest using autonomous under-canopy agricultural robots. Labor Optimization for Livestock Management: Audio monitoring can be combined with video to significantly improve individual tracking, behavior recognition, and other tasks for better livestock management. Thrust 2 researchers expanded on previously developed methods for separating mixed audio signal into constituent sources to predict the direction of audio sources by incorporating the scene geometry. Students from thrust 2 were able to apply these methods to the detection of COVID-19 infections with audio inputs, placing second in the worldwide DICOVA-II challenge. The thrust also developed three different pig and goat ethograms for behavior prediction with a variety of models. Environmental Resilience: The thrust created an image library containing >35,000 maize and sorghum leaf epidermal images, enabling the classification of common morphological features. Machine learning was used in combination with soil-canopy radiative transfer models for full use of proximal and airborne hyperspectral data for effective prediction of crop leaf and canopy level N status. In combination with the soil health thrust, these models were applied to predict SOC concentrations from proximal soil hyperspectral data. This result is in partnership with the soils thrust. Soil Monitoring and Health: In Year 2, the soils thrust was able to outfit two tile-drained fields in central Illinois with both wired and wireless sensor networks to measure input to output nutrient flux cycles. These fields led to the collection of 16,000 soil and plant measurements. With the rich dataset collected in Year 2, and existing datasets, novel methods based on graph neural networks were developed to predict soil N and P fluxes, which outperform previous supervised and semi-supervised learning methods. A new soil microbiome project launched and has begun with the collection of deep soil cores for in vitro microbiome analyses. Technology Adoption and Public Policy: In Year 2, thrust 5 extended their previous (static) weed ecology model with dynamic features to develop the Integrated Weed Economic-Ecology Dynamic (IWEED) model, to quantify yield loss caused by weeds depending on several static and dynamic input factors. Thrust members also established a choice experiment survey, sent to 10,000 US corn and soybean farmers, to examine factors driving willingness to plant cover crops with or without autonomous cover cropping robots. Education and Outreach: In Year 2, the education and outreach thrust has successfully hosted a K-12 computer science education summit, an REU summer intern program, CS-teacher endorsement training, and an AI foundry for agriculture applications summer workshop. In addition to these many programs, AIFARMS and the UIUC Center for Digital Agriculture received final Illinois state approval for a Master of Engineering in Digital Agriculture degree, expected to launch in Spring 2023.
Publications
- Type:
Journal Articles
Status:
Awaiting Publication
Year Published:
2022
Citation:
Cisneros-Velarde, P., Lyu, B., Koyejo, S. and Kolar, M. 2022. One Policy is Enough: Parallel Exploration with a Single Policy is Minimax Optimal for Reward-Free Reinforcement Learning, Arxiv, pre-print. https://arxiv.org/pdf/2205.15891.pdf.
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Kamtikar, S., Marri, S., Walt, B., Uppalapati, K. N., Krishnan, G. and Chowdhary, G. 2022. Visual Servoing for Pose Control of Soft Continuum Arm in a Structured Environment, IEEE Robotics and Automation Letters, 7(2). https://doi.org/5504-5511.10.1109/LRA.2022.3155821.
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Khanna, M., Atallah, S. S., Kar, S., Sharma, B., Wu, L., Yu, C., Chowdhary, G., Soman, C. and Guan, K. 2022. Digital Transformation for a Sustainable Agriculture in the United States: Opportunities and Challenges. Agricultural Economics, 00, 1-14.?https://doi.org/10.1111/agec.12733.
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Khanna, M. and Miao, R. 2022. Inducing the Adoption of Emerging Technologies for Sustainable Intensification of Food and Renewable Energy Production: Insights from Applied Economics. Australian Journal of Agricultural and Resource Economics, 66, 1-23. https://doi.org/10.1111/1467-8489.12461.
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Lai, T., Ji,. H. and Zhai, C.X. 2022. Improving Candidate Retrieval with Entity Profile Generation for Wikidata Entity Linking. Findings of the Association for Computational Linguistics, 36963711. https://doi.org/10.48550/arXiv.2202.13404.
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Wang, S., Guan, K., Zhang, C., Lee, D., Margenot, A.J., Ge, Y., Peng, J., Zhou, W., Zhou, Q. and Huang, Y. 2022. Using Soil Library Hyperspectral Reflectance and Machine Learning to Predict Soil Organic Carbon: Assessing Potential of Airborne and Spaceborne Optical Soil Sensing. Remote Sensing of Environment, 271, 112914. https://doi.org/10.1016/j.rse.2022.112914.
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Wu, J. and He, J. 2022. A Unified Meta-Learning Framework for Dynamic Transfer Learning. Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence, 3573-3579. https://doi.org/10.24963/ijcai.2022/496.
- Type:
Journal Articles
Status:
Published
Year Published:
2022
Citation:
Bernard, G. C., Bolden-Tiller, O., Egnin, M., Bonsi, C., McKinstry, A., Landon, Z., Chen, Y. Y., Ritte, I., Archie, T., Shafait, M.D., Chowdhury, G., Charleston, C., Turner A., Brown, A., Idehen, O., Mitchell, I., Boone, J., Peterson C. and Lockett, A. 2022. The Use of Autonomous Robots to Address Labor Demands and Improve Efficacy in Agriculture, COJ Rob Artificial Intellgence, 1(5). https://crimsonpublishers.com/cojra/pdf/COJRA.000523.pdf.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Lai, T., Ji, H. and Zhai, C. 2021. BERT Might be Overkill: A Tiny But Effective Biomedical Entity Linker Based on Residual Convolutional Neural Networks. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. 1631-1639. https://doi.org/10.18653/v1/2021.findings-emnlp.140.
|
Progress 09/01/20 to 08/31/21
Outputs
Target Audience:Research in AIFARMS has been disseminated to various target audiences;we have published several papers in peer reviewed journals and made presentations at professional meetings attended by other academic scientists. Through their engagement in AIFARMS research, undergraduate and graduate students as well as postdoctoral fellows are receiving training. AIFARMS team has also been reaching farmers, Extension agents and industry through podcasts, demonstrations and field days. We have also targeted the public as an audience for our activities through appearances on the local TV and at the Farm Progress Show. We are collaborating with several companies with an interest in our research to demonstrate and test practical applications and adapt our approaches to meet their needs. Changes/Problems:COVID-19 has strongly affected our capacity to meet in person. This inconvenience was quickly though partially overcome by regular online meetings within and between Thrusts, as well as between institutions and leadership teams. This problem also affected our ability to hold farmer surveys and focus groups that are important for the Technology Adoption research in Thrust 5, with only one small focus group possible in Year 1 and effectively delaying additional activities to Year 2. Also, as a result of COVID-19, students and staff could not travel together in one car. We mitigated this inconvenience by adding additional cars such that personnel could maintain social distancing. COVID-19 also prevented us from organizing our annual AIFARMS conference in the first year. Rather than organize a virtual meeting, we decided to cancel the event and hope to organize one in the middle or end of Year 2. The Advisory Board meeting and the NIFA and NSF funding agency reviews were originally intended to occur in person in conjunction with the annual conference, but were instead organized as virtual events via videoconference. COVID-19 has also affected the capacity to order and deliver instrumentation in times when needed (e.g. sensors in soil thrust). Alternative strategies were implemented by collecting data from other sources/type of sensors (e.g. drones collecting data on plant reflectance for nutrient content to compensate the lack of soil sensors). Another significant delay caused by the pandemic has been in hiring. The University of Illinois imposed a broad hiring freeze for many months in 2020, preventing or delaying the hiring of key staff members, including the Executive Director and the Assistant Director for Education and Outreach(both searches are well under way, and we expect them to complete soon). As for things to improve, we aim to have a stronger integration between Thrusts and between AI and crop modeling tools as data collection progresses. What opportunities for training and professional development has the project provided?In addition to the items illustrated in various Thrust Reports and the "Products" and "Other Products" sections, we would like to highlight the following opportunities provided by AIFARMS: Students and post-docs: Thirty-six (36) students and post-doctoral scientists work on AIFARMS projects. This includes nine (9) undergraduate students and five (5) post-doctoral scientists or visiting scholars. These students and post-docs are spread out across all of the AIFARMS thrusts and they are closely interactingin multidisciplinary fashion with AIFARM's soil scientists, plant scientists, animal scientists, computer scientists and sensor engineers. Post-docs and students are gaining valuable field experience (on farms or at animal facilities), and computer science students/post-docs have begun closely interacting with subject matter experts in soil science,animal science,genomics, andagronomy. Tenure-track faculty and senior researchers: Of the 42 senior researchers working at AIFARMS, 37 are tenure-track faculty or other senior researchers, spread out across the six thrusts. There is strong interaction among the tenure track faculty from disparate areas--soil science, animal science, computer sciences as joint projects have begun. This has enabled a new degree of necessary interdisciplinarity to help researchers involved in agriculture going forward. REU and undergraduate internships: In addition to post-graduate and graduate research, AIFARMS is also actively involved in the training of undergraduates. For instance, there were tenREU undergraduate interns embedded within AIFARMS projects. These tenstudents were all fully supported for two months during their summer internships, and participated in a number of AI+Agriculture research activities. They were also mentored in a wide range of career development goals, including the graduate school application process. Monthly seminars: Information is also disseminated via monthly seminars across AIFARMS and individual thrust meetings which are held weekly or once-in two weeks for each thrust. These thrust meetings are often attended by students and post-docs who gain valuable experience in the running of complicated, multi-faceted projects. Outreach: Outreach impacts training and professional development in both directions. This year there have been several outreach efforts with AIFARMS researchers participating in farmer focus groups, livestock farm visits, and related events. These interactions have helped the training of farmers to techniques developed by AIFARMS researchers and helped AIFARMS researchers' professional development from insights gained from non-academic customer environments. Joint research with industry:Research and serious technical interaction begun with Microsoft researchers, Windy City Labs engineers, and Terramar engineers has helped the professional development of AIFARMS members. Recruiting diverse students: We are working through the National Society of Minorities in Agriculture, Natural Resources and Related Sciences (MANRRS), to which TU and the University of Illinoishave strong affiliations, including the President-Elect who is a member of the AIFARMS leadership team. MANRRS will serve as a source of recruiting students for the project as well as provide social support and professional development opportunities for said participants. A subset of AIFARMS RA positions, with support from industry partners, will be dedicated to support those with Excluded Identities. iCAN program: This outreach program will address both the talent gap in tech and the underrepresentation of these groups in CS by providing foundational training in CS to agriculture students without said training at Illinois and partner institutions interested in professional development or obtaining a graduate degree in the Digital Agriculture program, thus bringing more students with agriculture backgrounds into computing. How have the results been disseminated to communities of interest?Results have been disseminated to academic communities of interest through conference presentations and publications in a broad range of disciplinary and cross-disciplinary venues. For example, the researchers on the team have published at least 14 papers in top-tier AI conferences, including NeurIPS, CVPR, ICCV, KDD and others. Research associated with Thrusts 3 and 4 was published in premier journals such as Nature Food and Nature Communications. These interdisciplinary journals have diverse audiences in agricultural sciences, engineering and computer sciences. Other research was published in leading economics journals. A few members of the AIFARMS team gave prestigious keynote presentations at conferences, including at the North American Plant Phenotyping Network (NAPPN) Annual Conference, and at SPLASH 2020, theACM SIGPLAN conference on Systems, Programming, Languages, and Applications: Software for Humanity, a premier conference in programming languages. The AIFARMS team also made external presentations to diverse audiences interested in high performance computing, programming systems, computer vision, digital agriculture, plant phenotyping, data science, robotics, agricultural economics and livestock welfare, amongst others. See the Products section for a list of all publications and presentations. Dr. Bruno Basso (MSU) organized and chaired a workshop for the National Academies of Science, Engineering and Medicine on "Exploring a Dynamic Soil Information System." The aim of the workshop was to examine how soil resources nationally might be dynamically and accurately monitored towardseveral important goals for soil health. The high-level focus of the workshop reflects several of the science and technology priorities identified in the 2016 Framework for a Federal Strategic Plan for Soil Science. A report from this workshop is publicly available (see Products). Dr. Sanmi Koyejo organized and chaired a workshop funded by the NSF HDR TRIPODS program on "Asymptotics and Non-Asymptotics in Control and Reinforcement Learning." This foundational AI research topic is important and highly relevant to AIFARMS because Reinforcement learning (RL) is a highly active area of research, blending ideas and techniques from control, optimization, machine learning, and computer science. In fact, AIFARMS researchers are exploring RL techniques for problems as diverse as controlling autonomous mobile robots and extracting new spectral patterns from large, but underused hyperspectral (phenotyping) data sets. Dr. Alex Schwing co-organized a workshop at CVPR 2021 on "3D Scene Understanding for Vision, Graphics, and Robotics." This foundational AI research area is critical to multiple projects within AIFARMS, a few of which include reconstructing 3D plant structure from images despite extensive occlusion of internal details, recognizing animal behaviors and human-animal interactions in livestock farms, and recognizing the surroundings in autonomous robot navigation in fields. We have also built our internal community within AIFARMS with a monthly seminar series, allowing AIFARMS scientists to present their research to the whole team and to build cooperation during a year when travel was limited. Finally, we reached the agricultural community by participating at the Farm Progress show, the University of Illinois Agronomy Day demonstration and through specific focus groups with growers. For more examples please reference "Products" and "Other Products". What do you plan to do during the next reporting period to accomplish the goals?In this section, we focus on the broad plans for Year 2 spanning the overall project. AIFARMS has made significant progress in the first year, and we will continue to strive for even more synergy between algorithmic, computational and practical agricultural technology within our Thrusts. Individual academic groups in the different disciplines contained within each Thrust have already begun to interact synergistically in Year 1, despite the limitations imposed on launching a broad new Institute in the midst of the pandemic. As hopefully more interaction becomes possible, we anticipate that the number of productive interactions between teams will increase noticeably in the second year. We also expect our productivity on all goals to accelerate in Year 2 as the project becomes fully staffed. In particular, we anticipate that graduate students newly hired during Year 1 will start to become productive, and that more educational programs will come online as degrees become approved. In our core technology goals, for example the small data problem or computer vision, researchers tend to be extremely busy and new students take some time before they can make significant contributions. In many of the projects requiring physical agricultural research infrastructure, delays are inevitable before the facilities can be fully functional. The core education and infrastructure parts of the project will be leveraged to speed both student training and facility performance. Thus, outputs are expected to increase rapidly over Years 2 and 3. More specifically, the leadership team of AIFARMS has several broad goals for the second year: Providing streamlined and regular scientific meetings and other interaction opportunities: A new monthly Distinguished Lecture Series, co-organized with the Center for Digital Agriculture, will bring in external speakers in a wide range of topics related to AI in digital agriculture. We will continue our regular, internal monthly all-hands team meeting where AIFARMS members present their research. Driving increased interactions between subprojects and even across thrusts: One example is to leverage the FedSSL software framework being developed in subproject 4 of Thrust 1 for more easily training machine learning models in berry picking (subproject 2 of Thrust 1), stomatal conductance in Thrust 3, and video monitoring of pigs in Thrust 4. Another example is expanding the interaction between Thrusts 3 and 4 on soil carbon measurement and prediction. Matchmaking between scientists with complementary expertise to solve specific problems. As one example, we will bring online one or two more subprojects to address newly-identified priorities in agricultural AI research. Encourage more subprojects within the team to make data, software and tools available publicly: Several subprojects are developing data sets (e.g., soils, pig videos) and software tools (e.g., FedSSL for semi-supervised learning, and the AVAT or Animal Video Analysis Tool for annotating videos of livestock), which could hold great value for the broader research community. The leadership, through the Data Management Working Group, will work to facilitate making these data sets and software tools available publicly, likely through the AIFARMS Github account. Ramping up the capacity and usage of our experimental facilities, especially the Illinois Autonomous Farm: Much of the robotics, edge computing, and sensing research takes place on this facility. The growth of IAF in Year 1, together with the new funding sources obtained in the first year, will support an expanded range of synergistic projects, all of which will contribute to strengthening the outcomes and impact from AIFARMS. Obtaining funding to expand and grow the REU program: We aim to support more students in Summer 2022, and hopefully also accommodate students from other BIPOC-serving institutions. All students in the program were fully supported last summer, and the leadership team will pursue funding opportunities to enable such an expansion while supporting all the students next year as well. Organizing in-person meetings and conferences: We anticipate that in-person events will become possible in Year 2, enabling us to organize such events. We are particularly keen to organize our in-person annual conference, after having to cancel that event for Year 1. Such events will enhance the educational experiences of our students as well as the quality of scientific communication within the team. Exploring synergy in the next year between AIFARMS and other large, newly-funded projects at Illinois over the past year: The NSF Center for Research On Programmable Plant Systems (CROPPS) may benefit from the AI, Machine Learning, Autonomous Farming, and Genomics research happening in AIFARMS, and conversely, AIFARMS may benefit from the new directions in plant breeding, programmable plants as sensors, and "Internet of Living Things" (IoLT) goals of CROPPS. The IBM-Illinois Discovery Accelerator Institute addresses a number of research areas with specific overlapping goals, including edge computing, artificial intelligence, carbon accounting, and carbon sequestration.
Impacts
What was accomplished under these goals?
In its first year of operation, the AIFARMS Institute has laid a strong foundation for a broad and impactful five-year research program. The Strategic and Implementation Plan describes the vision, the key research, education and outreach thrusts, the organizational structure, and other important planning activities that are key to laying such a foundation. The twentysubprojects within fiveresearch thrusts all bring together AI and agriculture researchers in close collaborative efforts to explore how foundational advances in AI can impact important challenges in agriculture. The Education and Outreach thrust is contributing to important, impactful efforts for inspiring the nextgeneration to explore digital agriculture and to help grow a skilled digital agriculture workforce. A successful in-person REU program gave a number of students hands-on research experience as well as wide-ranging career mentoring. New external research funding sources obtained by team members for closely related and complementary research significantly expands our scope and potential for impact. Multiple new industry engagements launched this year or currently under discussion have the potential to translate technical innovations from the group's research into commercial products. Team members have been invited to speak at many prominent research and education events, and organized several valuable community events. More specifically, the major accomplishments of the AIFARMS Institute in the first year are as follows (a few of the most significant accomplishments are marked with §). Accomplishments in Foundational AI Research We have identified six "foundational AI research challenges," which arise repeatedly in many different domains, including digital agriculture, and each of these impacts several different subprojects within AIFARMS. These foundational challenges are described briefly in the Strategic Plan. During the first year, we have made significant progress on all these six problems. We have also tackled some fundamental challenges in Computer Vision, which is a critical technology in digital agriculture. Our accomplishments in these foundational areas are as follows. Advances on the Small Data Problem: We advanced algorithms for rapid adaptation of neural networks for control with few data points (Havens and Chowdhary, 2021). We also developed datasets and problem formulations for semi-supervised learning for identification of berries and obtained preliminary results. §We developed techniques based on continuous transfer learning, in which we take as input both a static source domain with abundant labeled data, as well as a time-evolving unlabeled target domain, and output a predictive model for the target domain at the current time stamp. Our proposed method is based on a novel label-informed domain discrepancy measure, and it effectively integrates a variational autoencoder for feature extraction. §In another strategy based on "indirect invisible poisoning attacks," we take as input a base algorithm, labeled source data, as well as unlabeled target data, and output poisoned data for the source domain that degrades the predictive performance on the target domain. Our proposed method is based on bi-level optimization, where the objective function uses the label-informed domain discrepancy in the source domain. Foundational Advances in Computer Vision: Novel algorithms for understanding object movement in video data using multi-object tracking and segmentation (MOTS). §Fundamental activity recognition technology in computer vision, developed so far using human dance but also applicable to livestock. Models human activity as a hierarchical process which is consistent with how experts view it. Our model spans low levels (raw images, image sequences), to mid-levels (human poses and body part movements), to high levels (sequences of actions that form semantic units) of human activity. We expect to extend this capability to the case of pigs, instead of humans. New Techniques for Heterogeneous Data Fusion : Two of the techniques above for the "small data" problem - continuous transfer learning, and indirect invisible poisoning attacks - also provide novel benefits for learning from heterogeneous data sources. §Another new approach to heterogeneous data fusion uses a novel model called Network of Tensor Time Series (NeT3), which includes separate models to incorporate explicit relationship networks of the time series and model the implicit relationships among co-evolving time series. Advances in Learning for Control: §We have significantly improved the reliability of small robot navigation amongst corn and soybean fields using visual control learning, replacing expensive LIDAR and GPS sensors with far less expensive consumer video cameras, supported by recent advances in Convolutional Neural Networks (CNNs) for distance and heading (angle) estimation from a live video stream. (Also contributes to the foundational challenge of edge computing for machine learning.) We demonstrated through extensive field experiments that recent advances in CNN model pruning can dramatically reduce the computational requirements for mobile robot navigation without sacrificing navigation reliability. These optimizations serve to bring down the performance, energy requirements, and even the cost of computer hardware for robots by an order of magnitude or more. Strategies for Integrating Domain Knowledge into ML: Work on deep learning for control showed that it is possible to create hybrid reinforcement learning controllers that use approximate models of the system to guarantee stability (Narenthiran et al, 2021) In activity recognition, we use prior knowledge about constraints on parameters of motion that apply to humans to improve the accuracy of the generic models used for the predictions. Further, we expect to generalize the capability to separate the domain-dependent and domain-independent components. New techniques for Human Interaction with Autonomous Systems : §We developed natural language learning algorithms that learned keywords from WikiHow webpages for entity discovery and schema induction for procedural knowledge (Wang et al, 2021), to help answer gardening questions.
Publications
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2021
Citation:
1. National Academies of Sciences, Engineering, and Medicine. Exploring a Dynamic Soil Information System: Proceedings of a Workshop. Washington, DC: The National Academies Press, 2021. https://doi.org/10.17226/26170.
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
2. Basso, B. Precision conservation for a changing climate. Nat Food (2021). https://doi.org/10.1038/s43016-021-00283-z
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
3. Basso, B., Martinez-Feria, R.A., Rill, L. et al. Contrasting long-term temperature trends reveal minor changes in projected potential evapotranspiration in the US Midwest. Nat Commun 12, 1476 (2021). https://doi.org/10.1038/s41467-021-21763-7
- Type:
Journal Articles
Status:
Published
Year Published:
2020
Citation:
4. Khanna, M. Digital Transformation of the Agricultural Sector: Pathways, Drivers and Policy Implications Applied Economic Perspectives and Policy, October, 2020; https://doi.org/10.1002/aepp.13103
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
5. Khanna, M and R. Miao, Inducing the Adoption of Emerging Technologies for Sustainable Intensification of Food and Renewable Energy Production: Insights from Applied Economics Australian Journal of Agricultural and Resource Economics (revise and resubmit)
- Type:
Journal Articles
Status:
Under Review
Year Published:
2021
Citation:
6. Khanna, M., S. Atallah, S. Kar, B. Sharma, L. Wu, C. Yu, G. Chowdhary and C. Soman, Digital Transformation for a Sustainable Agriculture in the US: Opportunities and Challenges under preparation for submission to Agricultural Economics
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
7. A.Shirke,A. Saifuddin, A.Green-Miller, I. Condotta, A. Kotnana,O. Kocabalkanli, N.Ahuja, R. N.Dilger, and M. Caesar. "Tracking Grow-Finish Pigs Across Large Pens Using Multiple Cameras". AgEng2021.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2021
Citation:
8. A. Shirke, J. Li, A. Green-Miller, T. Williams, X. Hu, A. Luthra, N. Ahuja, M. Caesar. "Tracking Grow-Finish Pigs Across Large Pens Using Multiple Cameras".CV4Animals CVPR Workshop 2021. Accepted.
- Type:
Journal Articles
Status:
Submitted
Year Published:
2021
Citation:
9. Active Learning with Graph Neural Network Dynamics, submitted to NeurIPS 2021, in collaboration with Andrew, Supratik
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
10. J. Wu, J. He. Indirect Invisible Poisoning Attacks on Domain Adaptation. KDD 2021
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
11. B. Jing, H. Tong, Y. Zhu. Network of Tensor Time Series. WWW 2021
- Type:
Journal Articles
Status:
Published
Year Published:
2020
Citation:
12. I.-J. Liu, R. Yeh, A.G. Schwing. High-Throughput Synchronous RL. NeurIPS 2020
- Type:
Journal Articles
Status:
Published
Year Published:
2020
Citation:
13. Z. Ren, R. Yeh, A.G. Schwing. Not all unlabeled data are equal: Learning to weight data in semi-supervised learning. NeurIPS 2020
- Type:
Journal Articles
Status:
Published
Year Published:
2020
Citation:
14. R. Sun, T. Fang, A.G. Schwing. Towards a better global loss landscape of GANs. NeurIPS 2020
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
15. P. Zhuang, O. Koyejo, A.G. Schwing. Enjoy your Editing: Controllable GANs for Image Editing via Latent Space Navigation. ICLR 2021
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
16. Z. Ren, I. Misra, A.G. Schwing, R. Girdhar. 3D Spatial Recognition without Spatially Labeled 3d. CVPR 2021
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
17. C. Graber, G. Tsai, M. Firman, G. Brostow, A.G. Schwing. Panoptic Segmentation Forecasting. CVPR 2021
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
18. Y.-T. Hu, J. Wang, R. Yeh, A.G. Schwing. SAIL-VOS 3D: A synthetic dataset and baselines for object detection and 3D mesh reconstruction from video data. CVPR 2021
- Type:
Journal Articles
Status:
Under Review
Year Published:
2021
Citation:
19. Wang, S., Guan, K., Ainsworth, E.A. et al. Airborne hyperspectral imaging of nitrogen deficiency on crop traits and yield by machine learning and radiative transfer modeling. International Journal of Applied Earth Observation and Geoinformation
- Type:
Journal Articles
Status:
Under Review
Year Published:
2021
Citation:
20. Zhou, Q., Wang, S., Guan, K. et al., High-performance atmospheric correction of airborne hyperspectral imaging spectroscopy: model intercomparison, parameter retrieval, and machine learning surrogates. Remote Sensing of Environment
- Type:
Journal Articles
Status:
Under Review
Year Published:
2021
Citation:
21. Wang, S., Guan, K., et al. Using soil library hyperspectral reflectance and machine learning to predict soil organic carbon: assessing potential of airborne and spaceborne optical soil sensing. Remote Sensing of Environment.
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
22. Baoyu Jing, Hanghang Tong, Yada Zhu: Network of Tensor Time Series. WWW 2021: 2425-2437
- Type:
Other
Status:
Under Review
Year Published:
2021
Citation:
23. Baoyu Jing, Si Zhang, Yada Zhu, Bin Peng, Kaiyu Guan, Hanghang Tong: iTime: Instance Guided Time Series Imputation
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2021
Citation:
24. N. Ahuja and N. Mahajan, ICCV 2021: Compact, Explainable Deep Learning models
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2021
Citation:
25. Williams, T. and A. Green-Miller. (2021). Engineered Resilience in Livestock for Improved Animal Welfare. Abstract Accepted for ASAS Conference (towards Journal of Animal Science)
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2021
Citation:
26. X. Hu and N. Ahuja, ICCV 2021, HumanPose Sequence Estimation and Recognition
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2021
Citation:
27. M. Chatterjee, A. Cherian, N. Ahuja, ICCV 2021: Audio-Visual Fusion
- Type:
Journal Articles
Status:
Published
Year Published:
2020
Citation:
28. Basso, B. & Antle, J. Nature Sustainability. 3, 254256 (2020)
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
29. Basso B (2021) Precision conservation for a changing climate. Nature Food 2: 322-323
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
30. Basso B, et al., (2021) Contrasting long-term temperature trends reveal minor changes in projected potential evapotranspiration in the US Midwest. Nature Communications 12: 1476
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
31. Maestrini B, Basso B (2021) Subfield crop yields and temporal stability in thousands of US Midwest fields. Precision Agriculture DOI: 10.1007/s11119-021-09810-1.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2021
Citation:
32. Northup, D, Basso. B et al., 2021. Novel Technologies for Emission Reduction Complement Conservation Agriculture To Achieve Negative Emissions From Row Crop Production. Proceedings National Academy of Sciences, Vol. 118 No. 28
- Type:
Journal Articles
Status:
Published
Year Published:
2021
Citation:
33. Jiayang Xie, Samuel B Fernandes, Dustin Mayfield-Jones, Gorka Erice, Min Choi, Alexander E Lipka, Andrew D B Leakey, Optical topometry and machine learning to rapidly phenotype stomatal patterning traits for maize QTL mapping, Plant Physiology, 2021;, kiab299, https://doi.org/10.1093/plphys/kiab299
- Type:
Conference Papers and Presentations
Status:
Other
Year Published:
2021
Citation:
1. Khanna, M. Digital Transformation for a Sustainable Agriculture in the US: Opportunities and Challenges International Agricultural Economics Association Conference, August 29, 2021.
- Type:
Conference Papers and Presentations
Status:
Other
Year Published:
2021
Citation:
2. Khanna, M. Economic Incentives for Robotic Weed Control Cluster of Excellence PhenoRob Robotics and Phenotyping for Sustainable Crop Production at the University of Bonn, Sept.24, 2021.
- Type:
Conference Papers and Presentations
Status:
Other
Year Published:
2021
Citation:
3. Finger, R., Huber, R., Wang, Y and M. Khanna, Panel Discussion: Digital innovations for more sustainable agricultural landscapes , Landscape 2021, Berlin, September 20-22, 2021.
- Type:
Conference Papers and Presentations
Status:
Other
Year Published:
2021
Citation:
4. Elizabeth Ainsworth, Using hyperspectral reflectance to estimate and map photosynthesis in a soybean NAM population. Plenary Talk, North American Plant Phenotyping Network Annual Conference, February 2021 (online).
- Type:
Conference Papers and Presentations
Status:
Other
Year Published:
2021
Citation:
5. Tiffani Williams, The Art and Science of Black Farming, STEM Illinois Communiversity Think & Do Tank and Carle Illinois College of Medicine, February 27, 2021.
- Type:
Other
Status:
Other
Year Published:
2021
Citation:
6. Bruno Basso, US National Academy of Sciences, Engineering and Medicine, Reducing the Health Impacts of the Nitrogen Problem: An Environmental Health Matters Workshop, Digital Agriculture to Reduce Nitrogen Losses across the U.S. Corn Belt. Virtual meeting
- Type:
Other
Status:
Other
Year Published:
2021
Citation:
7. Bruno Basso, 2021 Columbia University, Integrating crop models, AI, and sensing for scaling sustainable agricultural systems
- Type:
Conference Papers and Presentations
Status:
Other
Year Published:
2021
Citation:
8. Bruno Basso, 2021 AgMIP annual meeting, Modeling Circular Agricultural Systems, Columbia University.
- Type:
Other
Status:
Other
Year Published:
2021
Citation:
9. Andrew Leakey, The Phenomics of Stomata and Water Use Efficiency in C4 crops (December 2020). ARPA-E TERRA Program PIs Meeting
- Type:
Other
Status:
Other
Year Published:
2021
Citation:
10. Andrew Leakey, The Phenomics of Stomata and Water Use Efficiency in C4 crops (October 2020). Martin and Ruth Massengale Lecture to the Annual Meeting of the Crop Science Society of America
- Type:
Other
Status:
Other
Year Published:
2021
Citation:
11. Andrew Leakey, The Phenomics of Stomata and Water Use Efficiency in C4 crops (Feb 2021). University of Missouri Interdisciplinary Plant Group seminar
- Type:
Other
Status:
Other
Year Published:
2021
Citation:
12. Andrew Leakey, The Phenomics of Stomata and Water Use Efficiency in C4 crops (March 2021). UIUC Department of Plant Biology colloquium
- Type:
Other
Status:
Other
Year Published:
2021
Citation:
13. Andrew Leakey, The Phenomics of Stomata and Water Use Efficiency in C4 crops (April 2021). DOE BRC Sorghum workshop
- Type:
Other
Status:
Other
Year Published:
2021
Citation:
14. Andrew Leakey, Overcoming bottlenecks in field-based root phenotyping using thousands of minirhizotrons (May 2021). 11th Symposium of the International Society of Root Research and Rooting 2021
- Type:
Other
Status:
Other
Year Published:
2021
Citation:
15. Andrew Leakey, Phenotyping stomatal anatomy and function (Sept 2021) Society for Experimental Biology Environmental Physiology Group, Virtual Workshop on Field and Laboratory Techniques.
- Type:
Other
Status:
Other
Year Published:
2021
Citation:
16. Alex Schwing, AIFARMS: Artificial Intelligence for Future Agricultural Resilience, Management and Sustainability, Vision for Agriculture Workshop at CVPR, 2021.
- Type:
Other
Status:
Other
Year Published:
2020
Citation:
17. Vikram Adve, AIFARMS: Artificial Intelligence for Future Agricultural Resilience, Management, and Sustainability, Washington State University Digital Agriculture Summit: How AI & Cyberinfrastructure are Impacting the Evolution of Digital Agriculture, Oct. 7, 2020.
- Type:
Other
Status:
Other
Year Published:
2020
Citation:
18. Vikram Adve, AIFARMS: Artificial Intelligence for Future Agricultural Resilience, Management, and Sustainability, SRC Center for Applications Driving Architectures (ADA) Fall Symposium, Nov. 10, 2020.
- Type:
Other
Status:
Other
Year Published:
2020
Citation:
19. Vikram Adve, Why Digital Agriculture is Fertile Ground for Software Systems Research, SPLASH 2020, the ACM SIGPLAN conference on Systems, Programming, Languages, and Applications: Software for Humanity (Keynote Presentation), Nov 19, 2020
- Type:
Other
Status:
Other
Year Published:
2021
Citation:
20. Vikram Adve, AI for Agricultural Innovation, Grainger College of Engineering AI Research Webinar, Feb. 12, 2021
- Type:
Other
Status:
Other
Year Published:
2021
Citation:
21. Vikram Adve, AIFARMS: Artificial Intelligence for Future Agricultural Resilience, Management, and Sustainability, Virginia Tech Center for Advanced Innovation in Agriculture (CAIA), March 23, 2021
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2021
Citation:
22. Vikram Adve, Computational Needs for the AIFARMS National AI Institute, Coalition for Academic Scientific Computation (CASC) Annual Spring Conference, April 7, 2021
- Type:
Other
Status:
Other
Year Published:
2021
Citation:
23. Vikram Adve, AIFARMS: Artificial Intelligence for Future Agricultural Resilience, Management, and Sustainability,, Digital/AI Seminar Series of the Centre for Bhutan Studies and GNH, June 3, 2021.
|
|