PeopleofSuperintelligence

Researcher

Kevin Swersky

Kevin Swersky is a machine-learning researcher whose work spans fair classification, Bayesian optimization, discrete sampling, accelerator design and generative video. His papers explored fair representations, efficient sampling, accelerator design, diffusion-model tuning and video reasoning. ICML recognized two of his coauthored papers with an Outstanding Paper Honorable Mention in 2021 and a Test of Time Award in 2023. As of October 2026, he is a research scientist at Google, with recent work on video reasoning and language-model planning.

Now at
Google
Works
88
Citations
17,277
h-index
29

Publication numbers from OpenAlex, as of 15 October 2026.

Timeline

  1. 2013

    Learning fair representations

    In 2013, he coauthors Learning Fair Representations with Rich Zemel, Yu Wu, Toni Pitassi and Cynthia Dwork. The paper presents a fair-classification algorithm with group and individual fairness.

  2. 2020

    Amortized Bayesian optimization

    In 2020, he coauthors Amortized Bayesian Optimization over Discrete Spaces. The paper trains a generative model to propose candidates that maximize the acquisition function.

  3. 2021–2022

    Discrete sampling and accelerator design

    In 2021, he coauthors Oops I Took A Gradient, a scalable approximate sampler for discrete distributions, and ICML gives the paper an Outstanding Paper Honorable Mention. In 2022, he coauthors Data-Driven Offline Optimization for Architecting Hardware Accelerators, an offline method for accelerator design.

  4. 2023

    A Test of Time Award

    In 2023, ICML gives Learning Fair Representations its Test of Time Award. Swersky is one of the five authors.

  5. 2024

    Fine-tuning diffusion models

    In 2024, he coauthors Directly Fine-Tuning Diffusion Models on Differentiable Rewards, which fine-tunes diffusion models to maximize differentiable rewards.

  6. 2025

    Reasoning with video models

    In 2025, he coauthors Video models are zero-shot learners and reasoners, which reports Veo 3 solving visual tasks such as object segmentation, edge detection and maze reasoning.

  7. As of October 2026

    Research scientist at Google

    In July 2026, he coauthors Google DeepMind research showing that visual prompt engineering improves video reasoning. In October 2026, he coauthors Analysis of Optimality of Large Language Models on Planning Problems for COLM. As of October 2026, he is a research scientist at Google.

Sources and credits

Photo: We have no published photo of Kevin Swersky yet. Send us a free-licensed photo.

  1. Learning Fair Representations proceedings.mlr.press. Checked 15 October 2026.
  2. Amortized Bayesian Optimization over Discrete Spaces proceedings.mlr.press. Checked 15 October 2026.
  3. ICML 2021 Awards icml.cc. Checked 15 October 2026.
  4. Kevin Swersky research.google. Checked 15 October 2026.
  5. ICML 2023 Awards icml.cc. Checked 15 October 2026.
  6. Directly Fine-Tuning Diffusion Models on Differentiable Rewards — Google DeepMind deepmind.google. Checked 15 October 2026.
  7. Video models are zero-shot learners and reasoners — Google DeepMind deepmind.google. Checked 15 October 2026.
  8. Visual prompt engineering for video models — Google DeepMind deepmind.google. Checked 15 October 2026.
  9. Google at COLM 2026 research.google. Checked 15 October 2026.
  10. Pre-trained Gaussian processes for Bayesian optimization — Google Research research.google. Checked 15 October 2026.