Skip to main content

ARS AI Innovation Fund (FY26)

Overview

The goal of the ARS Artificial Intelligence Center of Excellence (AI-COE) is to enable innovative ARS science by promoting the adoption and use of AI and machine learning (ML) tools and methods in agricultural research. The FY26 competition is now complete, and instructions specific to the FY27 call for proposals will be posted here in early FY27. For examples of successful proposal topics, please see the abstracts of the AI Innovation Fund proposals funded in FY2026, FY2024, FY2023, FY2022, and FY2021 (this program was paused for FY2025).

Proposal guidelines

Scope

Projects of high priority for funding are those that:

  1. Develop or adapt an AI/ML method that empowers ARS scientists to answer a specific question/problem or test a hypothesis of agricultural importance.
  2. Develop or adapt AI/ML technologies to create a prototype digital product that solves a need for producers or agricultural researchers.

Proposals should be primarily focused on developing, adapting, or applying methods that fall into the category of AI or ML (see definitions below). Please refer to the abstracts of the AI Innovation Fund proposals funded in FY2026, FY2024, FY2023, FY2022, and FY2021 for examples of successful proposal topics.

Researchers developing a method or digital product are encouraged to define a minimum viable product as a deliverable.

Successful projects should:

  1. Address real-world model concerns, such as data shift.
  2. Have AI/ML development and/or application of AI/ML methods to scientific research as a primary focus.
  3. Demonstrate that the project has a high probability of completion with impacts on an agricultural research question/problem.
  4. Utilize SCINet computing resources, including SCINet’s high-performance computing (HPC) clusters, Ceres and Atlas. These HPC systems are equipped with standard CPU nodes, graphics processing unit (GPU) nodes, and high-memory nodes. SCINet’s clusters support a wide range of modern AI/ML software and workflows, including industry-leading deep learning frameworks and containerization technologies.

Topics that will not be considered for funding

Training, workshop, and working group activities are not supported by this call. Please see the SCINet web site for ways to get involved in these activities or contact the SCINet Office with questions (ARS-SCINet-Office@usda.gov).

Project Funding

We expect to fund 4 to 6 proposals up to $100,000 each in each fiscal year. Funds must be spent or obligated in the same fiscal year in which they are awarded, which may require a collaborative agreement.

Proposal format and submission

All proposals must be submitted using the online submission form (TBA). The PI’s RL or supervisor must approve the proposal prior to submission (approval will be indicated on the submission form). A complete application will include:

  • A proposal abstract of no more than 1,500 characters.
  • Proposal (project description) of up to 2 pages in length that clearly lays out a specific challenge or question, proposes a method or tool to be developed or applied to solve the challenge or to answer the question, and demonstrates that the research team has the capability to complete the project. Deliverables for the project should be defined.
  • A detailed project budget provided as an Excel spreadsheet. Please use REE budget form 455.
  • A budget explanation of no more than 1,500 characters.

The proposal should be submitted as a PDF document with margins of no less than 1 inch and font size of no less than 11. The proposal may include figures, which should be included in the 2-page limit. References are not included in the 2-page limit for the proposal. Only one proposal as the lead investigator responsible for project completion can be submitted by a scientist, although a scientist can be a member of multiple proposal teams. We encourage teams of investigators collaborating on a problem.

Deadline for proposal submission: TBD

Eligibility: ARS Category 1, 4, or 6 scientists with RL or supervisor approval.

Definition of AI/ML technologies

(These are examples and not inclusive of all possible methods and tools.)

AI methods involve automated decision-making or inference from data and use methods from the subfields of:

  • machine learning (including deep learning)
  • generative statistical modeling
  • mathematical optimization (integer programming and operations research)
  • machine reasoning and logic programming
  • knowledge representation
  • recommender systems

Machine learning involves training a model with data and then making decisions or answering questions using that model. ML methods include:

  • tasks like classification, regression, dimensionality reduction, and clustering;
  • domain areas like natural language processing, computer vision, and time-series analyses;
  • methods like decision trees and random forests, neural networks (including deep learning), Bayesian networks, and support vector machines;
  • generative deep learning models such as autoregressive language models.