AI Science at Scale

  • AI Science at Scale
  • Collaboration
  • View Collaboration
  • Director’s Colloquium
  • Student Programs
  • Postdoctoral Research
  • Workforce Programs Office (WFPO)
  • National Security Education Center
  • National Security Life Sciences – Biosecurity Preparedness
  • Hydrogen Portal
  • Fellows
  • Feynman Center For Innovation
  • Internships
  • Nuclear Security and Safeguards Training
  • Center for National Security and International Studies
  • Pathogen Research Databases
  • Research Opportunities
  • Opportunity for University of California System Researchers and National Laboratory Collaborators to Advance AI at Scale

    The University of California (UC) System, in partnership with NNSA laboratories, is seeking proposals from UC researchers for collaborative projects that advance the frontiers of Artificial Intelligence and Machine Learning at scale.

    Our central goal is to develop scientific AI systems that benefit from computational scale while also harnessing the power of interdisciplinary AI and science expertise. These efforts – scaled up by a focus on both large AI models and cross-cutting teams – should operate seamlessly across multiple campuses while deeply integrating with National Laboratories. These efforts must address critical challenges in national security, scientific discovery, and technological innovation – and must do so at scale. 

    By building robust, interdisciplinary teams, we aim to leverage novel AI techniques, computational resources, and domain expertise to develop scalable solutions that significantly advance key application areas, including Multiphysics Applications, Biological Systems, and Materials Discovery.

    Teams should focus both on developing exciting new AI Capabilities for Science and on driving forward a single specific Technical Focus Area.

    Workshop

    Wednesday, October 16, 2024

    The University of California, Riverside, is hosting a one-day Information Workshop as part of this call.  At this workshop, we will have plenary sessions that will present the objectives of the three key technical focus areas: Biological Systems, Multiphysics Applications, and Materials Discovery.  We will also have breakout sessions in each focus area to allows the UC and National Laboratory colleagues to exchange ideas and begin to network and form teams in preparation for the proposals.

    Deadline Oct 4 2024; please register early due to facility limitations

    The UC Office of President has provided some support to cover expenses (lodging and transportation).  If you need financial support, please contact Gerardo Rodriguez of UC Riverside.

    If you are interested in engaging your colleagues and submitting a proposal as part of a team but are not able to attend the workshop, we encourage you to submit your “Letter of Intent.”  Please indicate which technical focus area you are interested in and include a short (1 paragraph) description of your technical interest. 

    PROPOSALS

    Successful proposals will deliver a cohesive AI ecosystem that drives forward a single, focused application area.  Projects should develop a unified interdisciplinary team with AI researchers and subject matter experts that jointly create a scientific AI system for the specific application built around novel AI contributions. Ideally, the proposed AI methods transcend the subject matter area and generalize to a wider range of problems, however the work and vision should be built specifically for only one of the application areas below. 

    Technical Focus Areas

    Objective: Develop AI-driven solutions to improve multiphysics simulation capabilities, with a focus on generating large-scale, validated AI models that can emulate complex physical phenomena. Examples include inertial confinement fusion, sea-ice for climate predictions, and others.

    Key Challenges:

    • Can large language models (LLMs), vision transformers (ViTs), or other architectures be adapted to create versatile emulators for multiphysics simulations?
    • How can simulations be effectively sampled to train these models with high reliability?
    • What role do physical and mathematical constraints play in the training process?

    Objective: Create foundation models that predict biological functions and evolutionary dynamics within complex ecosystems, and build new or purpose existing  molecules using AI-driven design and synthesis techniques, including as drug treatments.

    Key Challenges:

    • How can LLMs be used to predict nucleic acid and protein sequence functions or model the evolution of organisms in dynamic ecological systems?
    • Can AI models integrate vast amounts of genomic data to predict biological behaviors at scale?
    • What AI techniques can search for the best candidate small molecule compounds that will bind to selected cell targets to treat disease conditions, factoring in protein-protein interactions and protein conformational dynamics?
    • What AI techniques can be developed or adapted to design and synthesize new molecules, ensuring they possess desired functions or properties?
    • How can empirical data from experimental systems be effectively used to refine and validate AI-driven molecular design?

    Objective: Apply AI/ML techniques to the inverse design of materials and their experimental realization, focusing on optimizing properties such as mechanical response or properties relevant for the production of materials and their application to clean tech.

    Key Challenges:

    • Can AI models be developed to predict material properties at multiple scales (from molecular to continuum) and optimize material design accordingly?
    • How can multimodal data from experiments and simulations be leveraged to improve material design processes?
    • How can general science knowledge and AI models be integrated into self-driving laboratories

    Elements of a Coherent Ecosystem

    Proposals should not only aim to advance specific application area(s) but also contribute to strengthening the broader AI ecosystem that establishes partnerships spanning academia and national laboratories. The Frontiers of Artificial Intelligence for Science, Security, and Technology (FASST) initiative emphasizes the importance of a holistic approach where work on applications is tightly integrated with and enhances three foundational elements: Data, Models, and Compute.

    • Integration with Applications: Proposals should focus on creating or leveraging innovative data methods that enhance the quality, availability, and usability of data across the AI ecosystem. This includes the development of new data representations, tokenization strategies, and data pipelines that facilitate the flow of information between experimental, simulated, and real-world data sources.
    • Strengthening the Ecosystem: By advancing data methodologies within your application area, your work should help establish robust data practices that benefit other areas of research. For instance, methods developed for handling large, multimodal datasets in one application should be applicable to other domains, thereby improving data consistency and accessibility across the ecosystem.

    • Integration with Applications: Proposals should aim to develop or adapt AI models that are not only suitable for the specific application but also contribute to the broader modeling framework within FASST. This includes creating scalable architectures, improving training processes, and ensuring model robustness and validity.
    • Strengthening the Ecosystem: Work on application-specific models should lead to advances that can be generalized or transferred to other contexts, thus enriching the overall modeling capability within the AI ecosystem. For example, techniques developed to ensure model accuracy and scalability in a biological application might be adapted to improve model performance in materials science or multiphysics simulation.

    • Integration with Applications: Proposals should consider the computational demands of AI at scale, proposing methods that optimize the use of national lab assets, cloud resources, and emerging hardware. This includes addressing challenges related to strong and weak scaling, parallelism, and computational efficiency.
    • Strengthening the Ecosystem: By developing compute strategies that address the specific needs of your application, your work should also contribute to the optimization of computational resources across the AI ecosystem. This could involve creating new algorithms or tools that improve the efficiency of AI workloads, which can then be used in other research areas to enhance overall computational capabilities.

    In designing your project, consider how your work in one application area (e.g., multiphysics, biology, or materials discovery) can simultaneously contribute to and benefit from advances in data, models, and compute. A successful proposal will demonstrate a clear strategy for closing the loop between these elements, ensuring that progress in one area supports and is supported by the others, ultimately leading to a more cohesive and powerful AI ecosystem.

    Solutions for the technical areas must build novel AI contributions to the scientific AI ecosystem. Application of existing or traditional ML methods to application areas will not be sufficient. Transformational methods might include, but are certainly not limited to:

    • AI surrogates with uncertainty quantification (UQ) and accelerators to optimize predictive accuracy and scalability of numerical tools. 
    • AI models trained with advanced data representations, new multimodal agents to integrate empirical and numerical data, and multitask models. 
    • AI reasoning systems for reliable multiscale and multidomain predictions, and multiscale physics models combining first-principles and AI predictions. 
    • Experimental design with deep reinforcement learning, AI predictions connected to real-world facilities for validation, and closed design loops exploiting empirical data for continuous AI model refinement.
    • Mitigation of hallucinations from large AI models, expanded explainability of performance, enhanced reliability in the decision-making process (including quantifying uncertainty)
    • Theoretical contributions in all of the above.

    Submission Guidelines

    • Interdisciplinary Teams: Proposals must involve collaboration between AI/ML experts and domain-specific researchers. Teams should include faculty and students from relevant fields to ensure comprehensive coverage of the technical challenges.
    • AI Lifecycle: Approaches should demonstrate a closed-loop AI lifecycle, encompassing data preparation, training, validation, testing, and iterative improvement.
    • Meaningful Outcomes: The proposed research should aim to produce both a robust process/approach and a significant advance in the chosen application area.

    Funding and Resources

    Selected proposals will receive funding and access to state-of-the-art computational resources provided by the UC-managed national labs. Researchers will also have opportunities to collaborate with leading experts and contribute to a national-scale AI effort that supports both scientific discovery and national security objectives.

    Have questions?

    Need more information?