Kempner Institute Accelerator Awards

 Competitive proposals will:

• Address a clearly articulated research question
• Define specific aims, methods, and success criteria
• Establish technical readiness and feasibility within the award period, including a functioning, tested codebase with demonstrated performance and scalability relevant to the proposed work.
• Be structured as discrete projects, not exploratory or open-ended research programs

We welcome proposals in the following areas, provided they meet the criteria above: 

• Projects that elucidate the foundations of intelligence: mathematical and computational models of intelligence, cognitive theories of intelligence, and the neurobiological basis of intelligence.
• Projects which lead to new AI/ML methods or applications:  projects with clearly defined methodological, analytical, or performance advances. This includes new AI/ML methods with explicit evaluation benchmarks, scalable applications that demonstrate methodological novelty, and rigorous studies that generate insight into how AI/ML systems function. 

Projects which are focused on AI applications without a clear demonstration of either advancing novel AI/ML methods or demonstrate novel insight into the AI/ML function and performance of AI/ML algorithms will not be awarded. 

Applications that join the two primary areas of study (e.g., the study of biological intelligence with the development/understanding of new AI/ML methodologies) are of particular interest. This includes using insights from neuroscience to inspire new AI methods as well as using AI/ML to gain a deeper understanding of the brain. 

This is an in-kind award only, no cash funds, financial contributions, or associated fees will be provided. Proposals are expected to support discrete projects which can be completed within the allocated window. 

The application is not intended to support a laboratory’s research programs, exploratory projects that have not yet been tested, or for general laboratory use. 

• All applicants must have a primary, full-time faculty appointment at Harvard University and must be paid by, and run their research operation through, Harvard. The sole exception are faculty paid directly by HHMI who would otherwise meet all other criteria.
• Individuals with hospital-based appointments or research operations and/or non-profit, independent corporations are not eligible. The application is open only to ladder faculty members (assistant professors, associate professors, and professors) with principal investigator rights at a Harvard school.
• Kempner Institute Investigators and Associate Faculty are not eligible to apply. However, we encourage applications from Kempner Affiliate Faculty who otherwise meet the eligibility criteria. 
• Each application should designate one Principal Investigator (PI) as the Coordinating Principal Investigator. Applications that propose collaborative work may also designate a single additional PI (Co-PIs). The Coordinating PI will act as the administrative contact for the Kempner Institute . 
• Co-PIs must be eligible per the terms outlined above.  
• One Harvard: Project teams can include a mixture of faculty, student, staff, and/or postdoctoral fellows. All investigators and project team members listed on the application must have an active Harvard email address.
• All investigators and project team members must be eligible for an FAS Research Computing (FASRC) account and must have an active account in order to access the Kempner cluster. For project team members whose school is not an FARSC-affiliated  institution, we will make a good-faith effort to facilitate access. However, access for unaffiliated project team members cannot be guaranteed and is subject to FASRC policies and School approval.   
• Resource-allocation requests cannot exceed 64 GPUs. All requests must be supported by a clear and well-defined plan. Awarded allocations will be  strictly limited to what is necessary to achieve the specific aims described in the submitted and approved proposal. Use of allocated resources for work beyond that stated in the specific aims is not permitted.
• Projects must adhere to our Open Science Policies. All project artifacts, including any data, code, and model weights used or generated under the award must be made openly and freely available. 
• Projects must be ready-to-launch upon the receipt of award, therefore, proposals must demonstrate scalability up to a quarter of the requested GPUs. 

Award period and start date: Applications should be submitted for projects with a maximum duration of 30 days. Resource-allocation award period and start date will be dependent on the scope of the proposal and will be determined by the Kempner Institute and the grantee once the project is selected for an award. Most resource-allocations will likely begin within 90 days following notice of award.  

Resource Allocation: This grant provides access to the Kempner Institute’s AI cluster. This is an in-kind award—there are no cash funds, financial contributions, or associated fees provided with the award. 

Applicants may request a maximum of 64 GPUs, though smaller-scale projects are equally encouraged. Projects are expected to be completed within 30 days. 

The cluster is managed by FAS Research Computing and comprises 144 NVIDIA A100 (40GB) GPUs and 384 NVIDIA H100 (80GB) GPUs. 

Use of the Kempner cluster is strictly limited to the specific aims described in the submitted and approved proposal. Unauthorized use will result in immediate termination of access and permanent disqualification from future use of the cluster. 

End of Award: Upon the conclusion of the award, all awardees will be required to submit an end-of-award progress report. In addition, all awardees will be featured in the Deeper Learning research blog to share key findings with the broader research community. 

At the sole discretion of the selection committee, completed projects with successful results may be offered the opportunity to request  an additional period of access to pursue additional, related experiments.  

All applications must be completed and submitted through the Accelerator Awards Portal. The application consists of the following sections: 

• Coordinating PI Details: The information entered should be for the Coordinating Principal Investigator (Coordinating PI), who will be the person submitting the application on behalf of the team (if applicable). The Coordinating PI will take responsibility for managing the group collaboration (if applicable) and be the administrative point of contact for the Kempner Institute and any partners. Information about the Co-Principal Investigator(s) on the proposal should be entered where requested.
  • Name
  • Email
  • Degree(s)
  • Title/Position
  • Harvard Departmental affiliations
  • Link to professional website in URL format (e.g., http://mywebsite.com) 
  • Provide three publications that demonstrate your relevant expertise and directly support your ability to achieve the objectives outlined in this project proposal. 

• Co-Principal Investigator Details: The information entered should be for the listed Co-Principal Investigator (if applicable).
  • Name
  • Email
  • Degree(s)
  • Title/Position
  • Harvard Departmental affiliations
  • Justify the inclusion of the Co-PI by describing their specific contributions and the unique expertise they bring to fulfill the project’s goals.
  • Provide three publications for each Co-PI that demonstrates their relevant expertise and directly supports their ability to achieve their objectives in this project proposal.

• Project Proposal (500 words): The project proposal describes the project objectives and the potential impact of the proposed work. The proposal must include the following components:
  • Project Title
  • Proposal Body: Provide a summary of 500 words or less that must describe the following:
    • Describe how the project aligns with the Kempner research mission as described in (100 words). 
    • Describe planned project and key project objectives (250 words). 
    • Clearly define the specific scientific question you are addressing (50 words).
    • Describe the expected impact or innovation of the project on your field of study (100 words). 
  • Figures: (limit one-page) Figures summarizing preliminary data are not required but encouraged. Figure legends do not count towards total word count.
  • References Cited: (not included in word count) please include complete source references, if applicable.
  • Open Science Policy acknowledgement 
  • Blog post acknowledgement
  • Project team members: Provide a list of all project team members, their role on the project, and their Harvard email addresses. 

• Technical Readiness: Projects should be ready to launch upon receipt of award, therefore, proposals that have not demonstrated scalability up to at least a quarter of the requested GPUs will not be approved. Applicants will be required to demonstrate Technical Readiness that should include a report with the following components (please use the provided template):
  • Code Links: Provide the link(s) to your relevant GitHub repositories using a valid URL in format (e.g., https://github.com/{account}/{repository_name}). If your repository is currently private, please add the following users as collaborators: the Kempner Institute (GitHub username: KempnerInstitute).
  • Frameworks, Libraries, and Software Environment: Please specify the GPU-accelerated framework you intend to use (e.g. PyTorch, TensorFlow) along with any other AI/ML frameworks, software dependencies, or libraries that are required for your project. If the workflows are containerized, please list what containerization tools you are using. (Maximum of 200 words) 
  • External APIs and AI Services: Specify if your project would utilize any external APIs or AI services (e.g., OpenAI, Anthropic, or Hugging Face) and explain their specifications (e.g., API call frequency) and function within your workflow.
  • Scalability Testing: Please report system usage statistics for multi-GPU jobs when running the proposed workflow. We require testing to include a minimum of a quarter of the requested GPUs and ask that GPU type, GPU count, average peak memory per GPU, and average GPU utilization are reported. We also recommend sharing a SLURM job submission script configuration that has been tested on a high-performance computing cluster (although, not required). If you currently do not have access to an HPC cluster, there are resources available through the Harvard ecosystem such as kempner_requeue, public requeue partitions (FAS), and the Research Computing Core (Harvard Medical School) that allow for testing.
  • GPU Type: Specify the type of GPUs required (e.g., A100, H100 or mixed) and whether your workload depends on specific GPU features (e.g., low precision, such as FP8, or large VRAM).
  • Data Requirements: Describe the datasets you will be using. Please include their size and current storage location (e.g., AWS, on-premises, local devices). We recommend Globus (read more here) for faster file transfer. Please note that only data that can be made publicly available and rated data security level 1 or 2 is permitted. 
  • Storage Needs: How much storage will your project require for input data, models, container images, and other training artifacts? Do you anticipate that your model will generate large output files and checkpoints, if so, please estimate their size. Please round up to the nearest terabyte.  
  • Technical Expertise: Briefly describe the project team’s technical expertise in using GPUs, distributed systems, or large-scale AI/ML training.
  • Deliverables: List expected deliverables following the conclusion of the award (e.g., codebase, notebooks, datasets, models, weights, etc.). 

• Resource Planning Timeline: Upload your project timeline as a single page in PDF format. Please use the following to format your timeline: letter size pages, single-spaced, Arial font 11 point (or larger), 0.5 inch margins for all pages.
  • Project Phases, Timeline, and Resource Requirements: Over a maximum of 30 days, projects typically progress from smaller experimental stages to larger, more resource-intensive needs. Outline key phases of your projects within this timeframe, including milestones, evaluation checkpoints, refactoring time needed, number of iterations planned, and specify the number of GPUs required at each stage. For each stage, outline:
    1. the number and types of GPUs and CPU cores you will need 
    2. specify the duration and pace of resource usage (e.g. hours, days, weeks)
    3. describe how the scale of parameters or model complexity will evolve over time 
  • References Cited: (not included in word count) please include complete source references. 
  • Figures:  Figures summarizing the project phases/timeline are not required but encouraged. Figure legends do not count towards total word count. 

Awarded proposals will need to provide a brief project summary to be published on the Kempner Institute’s website. 

The Kempner Institute adheres to our core mission and values and open science policies in both the award selection and in the evaluation of progress. All application sections will be evaluated on scientific merit by a review committee. 

An expert review committee will evaluate proposals. All proposals will be scored based on scientific merit and alignment with the Institute’s mission. Proposals will also be evaluated for technical feasibility, and readiness to utilize advanced compute resources. 

Proposals that fail to meet the Open Science Policy or do not demonstrate readiness to scale will be disqualified.

Reviewers will evaluate proposals using the following four criteria:

1. Mission Alignment
2. Scientific Merit, Technical Innovation, and Impact
3. Technical Readiness  
4. Resource Planning Timeline 

Each section will be scored on a scale of 1 to 5:

• 1 = Poor – Major gaps or misalignment with the Institute’s mission.
• 2 = Fair – Some weaknesses; limited alignment or feasibility.
• 3 = Good – Solid, but with room for improvement.
• 4 = Very Good – Strong proposal, minor issues.
• 5 = Excellent – Outstanding in all respects; highly recommended. 

Project proposals will be reviewed beginning on May 25, 2026. Please note that at any time throughout the review process, committee members may reach out to applicants for clarification on any submitted application compo

All awardees will be required to submit an end-of-award progress report not to exceed 2-pages within 90 days of the completed project that includes:

• Key findings and potential impact on the field 
• How this award supported the mission and vision of the Kempner Institute. 
• A list of all presentations, manuscripts (submitted or in press), meetings or courses related to award activities. 
• A list of and links to any work products (e.g. code or data) created, developed, or curated using this award. 

In addition, all awardees will be required to prepare, in coordination with Kempner staff, a  Deeper Learning research blog to share key findings with the broader research community within 90 days of the completed project. 

All award recipients will be required to adhere to the Open Science Policy. This policy addresses areas such as research integrity, data access, intellectual property, developed software, non-software inventions, publication and collaboration. Projects that cannot adhere to the policy cannot be supported by this award, the institute, or its GPU cluster. 

We support full publication and public availability of research findings without conditions or restrictions on academic and publication freedom. All manuscripts that result from this award (in whole or in part) shall be submitted as preprints to bioRxiv, arXiv, or a similar service for sharing preprints, before or upon first submission to a journal. 

Any work supported by this award will need to acknowledge the Kempner Institute. We suggest the following wording: “This work has been possible in part by a gift from the Chan Zuckerberg Initiative Foundation to establish the Kempner Institute for the Study of Natural and Artificial Intelligence.” 

Upon the conclusion of this award, we ask for the full citation of any resulting publications. If the work has not been submitted for publication, we ask that we are notified immediately upon submission and acceptance.

Awarded proposals will be required to submit a publishable project summary to be published on the Institute’s website. Awardees will work in collaboration with the Kempner Institute in order to create and publish the project summary following Kempner blog submission guidelines.  Awarded compute resources are strictly limited to the specific aims described in the submitted and approved proposal. Any use outside the scope of the approved project is strictly prohibited. Unauthorized use will result in immediate termination of access and permanent disqualification from future use of the cluster. Usage may be monitored or audited at any time to ensure compliance.

Can I collaborate with Kempner Institute Research Fellows under this award?

We discourage those who already have access to the Kempner AI cluster to participate in this award given that the spirit of the award is to give access to those who are not affiliated with the Kempner Institute. However, we do encourage fellows to contribute intellectually.

Can I collaborate with Kempner Institute Investigators?

Kempner Institute Investigators or Associate Investigators are not eligible to participate in this award. Kempner Institute affiliate faculty can participate if they meet all other eligibility criteria. 

I do not have access to compute resources at the moment. How can I prove scalability? Is it really required?

It is required to prove scalability up to a quarter of the requested resources in order to be considered for an award. If you currently do not have access to an HPC cluster, there are resources available through the Harvard ecosystem such as kempner_requeue, public requeue partitions (FAS), and the Research Computing Core (Harvard Medical School) that allow for testing. Please contact FASRC (rchelp@rc.fas.harvard.edu) or HMSRC (rchelp@hms.harvard.edu) for access to these partitions. 

I can scale up to a quarter of the resources I’m requesting. However, my project is still in an exploratory phase. Can I still apply?

We encourage applications for projects that are ready-to-scale and have a clearly defined scientific question. Research in the exploratory phase is discouraged for this award. 

I meet all of the eligibility criteria but my school is listed as “unaffiliated” in the FASRC documentation. Can I still apply?

FASRC’s current list of schools that are “unaffiliated” is related to existing memorandums of understanding (MOUs). We will do our best to facilitate access and have an existing agreement with HMS. We do not, however, pay for storage and will grant access to a temporary scratch directory. If you require storage beyond the scratch directory system, you will need an MOU in place prior to the award period. If you are interested in establishing an MOU for your school please contact FASRC (rchelp@rc.fas.harvard.edu). 

Can I change or edit the team members listed in my application?

No. All team members listed in the application are the only team members that will be given access to the reservation.

I am an awardee but I would like some extra time on my reservation. Can I apply for an extension?

Completed projects with successful results may be offered the opportunity to request an additional period of access to pursue additional, related experiments. Please contact Janet Wallace (janet_wallace@harvard.edu) for more information.