Applications closed

Request for proposals on sample-efficient sequential Bayesian decision making


Bayesian optimization is a methodology for sample-efficient learning and optimization. By leveraging a probabilistic model, it allows practitioners and researchers to explore large design spaces using only a small number of experimental trials. At Facebook, we utilize Bayesian optimization to improve product experiences, infrastructure, and aid in cutting edge research. For example, Bayesian optimization may be used to learn personalized video playback algorithms that work well across a diverse set of devices and levels of connectivity. Machine learning teams like Instagram Feed & Stories Relevance use Bayesian optimization to refine their latest machine learning models through the use of online A/B tests. And teams at Facebook Reality Lab use Bayesian optimization to efficiently conduct research in the area of perception in only a fraction of the time that conventional experiments would require.

To enable and support this work, we developed and open-sourced BoTorch, a modular framework for Bayesian optimization research, and Ax, a turn-key framework for those who want to apply Bayesian optimization to their own problems. Our goal with BoTorch is to accelerate the pace of research in the area of Bayesian optimization and unlock new potential applications. With this RFP, we hope to deepen our ties to the academic research community by seeking out innovative ideas and applications of Bayesian optimization that further advance the field. We are committed to open source and will help awardees make the products of this RFP available to the public as part of BoTorch.

Facebook is pleased to invite faculty to respond to this call for research proposals. In order to support academic work that addresses our challenges and opportunities while producing generalizable knowledge, Facebook is pleased to offer two research awards of $50,000 and $25,000, respectively. Awards will be made as unrestricted gifts to the principal investigator’s host university. Awardees will be invited to present and engage in discussion with researchers at Facebook.

Award Recipients

Stanford University

Jeannette Bohg

University of California San Diego

Rose Yu

Applications Are Currently CLosed

Application Timeline

Launch Date

February 24, 2021


April 21, 2021

Winners Announced

June 2021

Areas of Interest

Areas of interest include, but are not limited to, the following:

  • Bayesian optimization in high-dimensional and structured search spaces: Some problems are parameterized by high-dimensional search spaces that may include discrete (ordinal and categorical) parameters. One challenge we face at Facebook is how to combine a number of different models that all interact with each other in some way, but whose interactions are unknown a priori and hard to evaluate, resulting in a combinatorial optimization problem.
  • Scaling Bayesian optimization to high-throughput, high-parallelization settings where hundreds or thousands of evaluations are run in parallel. Simulation optimization (e.g., for hardware design) with many available compute nodes is a setting where this becomes important. There are also important connections to multi-fidelity Bayesian optimization, since we may be able to run a large number of low-fidelity simulations at lower computational cost.
  • Novel applicationsof Bayesian optimization to the sciences, such as materials science, chemistry, optics, medicine, and perception research. We are excited to see how tooling such as BoTorch can allow cutting-edge Bayesian optimization methods to have impact outside their traditional application areas.
  • Preference learning, active learning, and real-time Bayesian optimization. In certain settings, we want to optimize a system with the human “in-the-loop,” but in lieu of specific objective metrics that rely on human feedback. Methods need to be fast enough to be able to do so in “real time,” i.e., so that the optimization loop can happen in a lab or remote setting without prolonged gaps between evaluations.
  • AutoML is a core application of Bayesian optimization, and we are interested in new techniques for improving its efficiency and robustness, i.e., methods that reduce both training costs and model size / prediction times, navigate the tradeoff between overall wall-time and computational cost (e.g., by allowing early termination of training loops), and are able to handle corrupted evaluations. Applications include hardware-software co-design, neural architecture search for generic deep learning models (in particular, methods that do not require modifications to the actual model and training code, which hence can be used with established infrastructure).
  • There are many interesting uses of probabilistic surrogate models and optimization beyond the typical Bayesian optimization setting. We are interested in problems that use more expressive surrogate models, such as Bayesian hierarchical models or Bayesian neural networks, for optimization, bandits, or related active learning tasks. We are especially interested in model classes and active learning tasks that are broadly useful, and are suitable for use with Monte Carlo acquisition functions.

Projects will be chosen based on the extent to which they address a problem of sufficiently broad interest, have achievable goals, and can be replicated or transported to other settings. We encourage applicants to consider using BoTorch in their projects and how inclusion of their work in BoTorch can benefit the community.


Proposals should include

  • A summary of the project (1-2 pages), in English, explaining the area of focus, a description of techniques, any relevant prior work, and a timeline with milestones and expected outcomes
  • A draft budget description (1 page) including an approximate cost of the award and explanation of how funds would be spent
  • Curriculum Vitae for all project participants
  • Organization details. This will include tax information and administrative contact details


  • Proposal must comply with applicable U.S. and international laws, regulations, and policies.
  • Applicants must be current full-time faculty at an accredited academic institution that awards research degrees to PhD students.
  • Applicants must be the Principal Investigator on any resulting award.
  • Facebook cannot consider proposals submitted, prepared, or to be carried out by individuals residing in or affiliated with an academic institution located in a country or territory subject to comprehensive U.S. trade sanctions.
  • Government officials (excluding faculty and staff of public universities, to the extent they may be considered government officials), political figures, and politically affiliated businesses (all as determined by Facebook in its sole discretion) are not eligible.

Frequently Asked Questions

Terms & Conditions

Facebook’s decisions will be final in all matters relating to Facebook RFP solicitations, including whether or not to grant an award and the interpretation of Facebook RFP Terms and Conditions. By submitting a proposal, applicants affirm that they have read and agree to these Terms and Conditions.

  • Facebook is authorized to evaluate proposals submitted under its RFPs, to consult with outside experts, as needed, in evaluating proposals, and to grant or deny awards using criteria determined by Facebook to be appropriate and at Facebook’s sole discretion. Facebook’s decisions will be final in all matters relating to its RFPs, and applicants agree not to challenge any such decisions.
  • Facebook will not be required to treat any part of a proposal as confidential or protected by copyright, and may use, edit, modify, copy, reproduce and distribute all or a portion of the proposal in any manner for the sole purposes of administering the Facebook RFP website and evaluating the contents of the proposal.
  • Personal data submitted with a proposal, including name, mailing address, phone number, and email address of the applicant and other named researchers in the proposal may be collected, processed, stored and otherwise used by Facebook for the purposes of administering Facebook’s RFP website, evaluating the contents of the proposal, and as otherwise provided under Facebook’s Privacy Policy.
  • Neither Facebook nor the applicant is obligated to enter into a business transaction as a result of the proposal submission. Facebook is under no obligation to review or consider the proposal.
  • Feedback provided in a proposal regarding Facebook products or services will not be treated as confidential or protected by copyright, and Facebook is free to use such feedback on an unrestricted basis with no compensation to the applicant. The submission of a proposal will not result in the transfer of ownership of any IP rights.
  • Applicants represent and warrant that they have authority to submit a proposal in connection with a Facebook RFP and to grant the rights set forth herein on behalf of their organization. All awards provided by Facebook in connection with this RFP shall be used only in accordance with applicable laws and shall not be used in any way, directly or indirectly, to facilitate any act that would constitute bribery or an illegal kickback, an illegal campaign contribution, or would otherwise violate any applicable anti-corruption or political activities law.
  • Awards granted in connection with RFP proposals will be subject to terms and conditions contained in the unrestricted gift agreement (or, in some cases, other mechanisms) pursuant to which the award funding will be provided. Applicants understand and acknowledge that they will need to agree to these terms and conditions to receive an award.