PeopleofSuperintelligence

Researcher

Kevin Scaman

Kevin Scaman is a machine-learning researcher whose work spans distributed optimization, graph theory, machine unlearning, and tabular foundation models. After earning engineering and MVA master's degrees at École Polytechnique and a 2016 PhD at ENS Paris-Saclay under N. Vayatis, he worked as a postdoctoral researcher at the Microsoft Research–Inria Joint Center in Saclay from 2017 to 2018. His coauthored paper on non-smooth distributed optimization won a NeurIPS 2018 Best Paper Award, and he led the optimization-for-machine-learning project team at Huawei Noah’s Ark in Paris from 2018 to 2021 before serving as a research scientist at Inria Paris and the DI ENS Argo team until June 2025. In 2025 and 2026, he coauthored ICML papers on machine-unlearning complexity and variance-reduced unlearning, and as of October 2026 he is an ML researcher at Fundamental focusing on tabular foundation models.

Now at
Fundamental
Badges

Timeline

  1. 2008–2016

    Engineering and machine-learning degrees

    He earns an engineering degree in applied mathematics at École Polytechnique from 2008 to 2011 and an M.Sc. in machine learning in the MVA program there from 2011 to 2012. At NIPS 2014, he coauthors “Tight Bounds for Influence in Diffusion Networks and Application to Bond Percolation and Epidemiology,” deriving theoretical bounds on the long-term influence of a node in an Independent Cascade Model. From 2013 to 2016, he completes a PhD in machine learning at ENS Paris-Saclay under the supervision of N. Vayatis with the thesis “Analysis and control of diffusion processes in networks.”

  2. 2017–2018

    Microsoft Research–Inria postdoc and MSDA

    From 2017 to 2018, he works as a postdoctoral researcher at the Microsoft Research–Inria Joint Center in Saclay. At ICML 2017, he coauthors “Optimal Algorithms for Smooth and Strongly Convex Distributed Optimization in Networks,” which introduces the multi-step dual accelerated (MSDA) method for gossip-based decentralized optimization.

  3. 2018

    A NeurIPS best paper

    With Francis Bach, Sébastien Bubeck, Yin Tat Lee, and Laurent Massoulié, he coauthors “Optimal Algorithms for Non-Smooth Distributed Optimization in Networks,” which wins a NeurIPS 2018 Best Paper Award. Under a local regularity assumption, the paper introduces the multi-step primal-dual (MSPD) method, an optimal first-order decentralized algorithm for non-smooth convex functions.

  4. 2018–2021

    Principal researcher at Huawei

    From 2018 to 2021, he serves as a principal researcher at Huawei Noah’s Ark in Paris and leads the lab’s optimization-for-machine-learning project team.

  5. 2021–2025

    Research at Inria and ENS

    He joins Inria Paris and the DI ENS Argo team as a research scientist in 2021, working across machine learning, optimization, and graph theory. In 2022, he begins a part-time associate professorship at École Polytechnique in Saclay, and his Inria role runs until June 2025. At AISTATS 2024 in May 2024, he coauthors “Minimax Excess Risk of First-Order Methods for Statistical Learning with Data-Dependent Oracles,” providing sharp upper and lower bounds on minimax excess risk for strongly convex and smooth statistical learning.

  6. 2025

    Machine unlearning complexity and Stab-SGD

    At ICML 2025, he coauthors “When to Forget? Complexity Trade-offs in Machine Unlearning,” which establishes upper and lower bounds on minimax computation times for machine unlearning. At NeurIPS 2025, he coauthors “Stab-SGD: Noise-Adaptivity in Smooth Optimization with Stability Ratios,” introducing the stability ratio of gradient estimates to measure local relative noise.

  7. 2026

    Variance-reduced unlearning and approximate backpropagation

    In February 2026, he coauthors “Unbiased Approximate Vector-Jacobian Products for Efficient Backpropagation,” replacing exact vector-Jacobian products with randomized unbiased approximations during backpropagation. At ICML 2026, he coauthors “Variance-Reduced (ε, δ)-Unlearning using Forget Set Gradients,” introducing the Variance-Reduced Unlearning (VRU) algorithm that incorporates forget-set gradients directly into its first-order update rule.

  8. As of October 2026

    ML researcher at Fundamental

    He is an ML researcher at Fundamental, where he focuses on extending tabular foundation models. In July 2026, he speaks on “What LLMs Learn (and Don’t) from Tables” at the ICML Foundation Models for Structured Data workshop.

Career

  1. NowFundamentalML researcher
  2. From 2022École PolytechniqueAssociate Professor
  3. 2021–2025InriaResearch scientist
  4. 2018–2021HuaweiPrincipal researcher
  5. 2017–2018Microsoft Research–Inria Joint CentrePostdoctoral researcher

Education

  1. 2016École Normale Supérieure Paris-SaclayPhD in machine learning
  2. 2012École PolytechniqueM.Sc. in machine learning
  3. 2011École PolytechniqueEngineering degree in applied mathematics

From the biography.

  1. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes Kevin Scaman. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other university. We compare doctorate and 5+ year employment links, career timing and current workplaces for 187 major-prize and 1,111 best-paper winners.
  2. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Huawei. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.
  3. People shaping AI work at a median of 3 organizations over a career: 3 for university starters and institute and lab starters, versus 2 for company startersIncludes Inria. Across 6,853 people shaping AI with a dated work role, we estimate a median of 3 parent organizations over a career (mean 3.6), comparing role groups, starting sectors and starting decades from the 1950s to 2020–2026.
  4. The median h-index is 48 at universities and 14 at frontier AI labsIncludes École Polytechnique. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

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

  1. Kevin Scaman - Personal Website kscaman.github.io. Checked 22 October 2026.
  2. Tight Bounds for Influence in Diffusion Networks and Application to Bond Percolation and Epidemiology papers.neurips.cc. Checked 22 October 2026.
  3. Publications | Kevin Scaman scaman.wordpress.com. Checked 22 October 2026.
  4. Optimal Algorithms for Smooth and Strongly Convex Distributed Optimization in Networks proceedings.mlr.press. Checked 22 October 2026.
  5. Awards neurips.cc. Checked 22 October 2026.
  6. [1806.00291] Optimal Algorithms for Non-Smooth Distributed Optimization in Networks arxiv.org. Checked 22 October 2026.
  7. ARGO - 2025 - Rapport annuel d'activité radar.inria.fr. Checked 22 October 2026.
  8. Minimax Excess Risk of First-Order Methods for Statistical Learning with Data-Dependent Oracles proceedings.mlr.press. Checked 22 October 2026.
  9. When to Forget? Complexity Trade-offs in Machine Unlearning proceedings.mlr.press. Checked 22 October 2026.
  10. NeurIPS Poster Stab-SGD: Noise-Adaptivity in Smooth Optimization with Stability Ratios neurips.cc. Checked 22 October 2026.
  11. [2602.14701] Unbiased Approximate Vector-Jacobian Products for Efficient Backpropagation arxiv.org. Checked 22 October 2026.
  12. Variance-Reduced $(\varepsilon, δ)-$Unlearning using Forget Set Gradients proceedings.mlr.press. Checked 22 October 2026.
  13. dotAI | Speak dotai.io. Checked 22 October 2026.
  14. Speakers | Foundation Models for Structured Data icml-structured-fm-workshop.github.io. Checked 22 October 2026.