Nicolas Berkouk

AI Scientific Expert - CNIL (French DPA) - AI Department

Nicolas Berkouk

I am currently an AI Scientific Expert at CNIL, the French Data Protection Authority. As part of the AI Department, I follow closely the scientific and technologic development of AI, in the light of actual (GDPR) and future (AI Act) European regulation. I conduct ML research project for CNIL, and collaborate with social scientists to understand certain aspects of the structure of the AI ecosystem.

Before CNIL, I obtained my Ph. D. from INRIA/École polytechnique in Topological Data Analysis under the supervision of Steve Oudot, and spent three years as a post-doctoral fellow at EPFL, in the Topology and Neuroscience team led by Kathryn Hess.

Broadly speaking, my academic research was focused on making abstract tools of algebraic topology such as derived sheaf theory amenable to understand data and algorithms. For an overview, you can have a look at my Ph.D. defense:

News

Our paper Singular Value Representation: A New Graph Perspective On Neural Networks with Dan Meller got accepted to AISTAT23! Watch the SVR of a fully connected NN learning to classify the MNIST dataset:

Interested in the interplay between Euler Calculus, Sheaf theory and interleaving distance ? Check my new paper here.

Upcoming events

Past events

  • August 2023 Invited speaker for the TDA week as part of ICIAM 2023, Kyoto, Japan.
  • December 2022 Interactions between representation theory and topological data analysis workshop, CAS, Oslo. Persistence and the Sheaf-Function Correspondence.
  • November 2022 Chaire Pari Seminar (Paris - France). The field of Explainable AI : designing machines to explain machines ?
  • November 2022 Online workshop on computational persistence. Projected Barcodes and Integral Sheaf Metrics
  • October 2022 ETH Symplectic Geometry Seminar (Zürich - Switzerland), K-theory of constructible sheaves and persistence.
  • July 2022 AI and Health : Interdisciplinary Approaches Colloquium, Lebesgue Center for Mathematics. The field of explainable AI : designing machines to explain machines ?
  • June 2022 ATMCS Conference 2022 (Oxford - UK), Projected Barcodes and Integral Sheaf Metrics.
  • June 2022 Oxford TDA Seminar, Projected Barcodes and Integral Sheaf Metrics.
  • 18th February 2022 COSTECH (UTC - Compiègne) Seminar. The field of Explainable AI : designing machines to explain machines ?
  • 17th February 2022 DataShape (INRIA - France) Seminar, Persformer: A Transformer Architecture for Topological Machine Learning, , joint with Raphael Reinauer.
  • 10th February 2022 AlToGeLiS online Seminar. Persformer: A Transformer Architecture for Topological Machine Learning, joint with Raphael Reinauer.
  • 10th February 2022 STS Lab (UNIL - Lausanne) Seminar. The field of Explainable AI : designing machines to explain machines ?
  • August 2021 Sheaves and Homotopical Methods for Topological Data Analysis, Mini-Symposium as part of the SIAM AG21' conference, Co-organizer with François Petit.
  • July 2021 5th Conference on Geometric Science of Information, Paris, France. Talk: Algebraic Homotopy Interleaving Distance.
  • May 2021 AMLD workshop on AI and Topology, EPFL, Lausanne. Talk: Level-Sets Persistence and Sheaf Theory.
  • April 2021 Workshop on Topological Data Analysis, IMSI, Chicago, US. Talk: Ephemeral Persistence Modules and Disance Comparisons.
  • June 2020 Applied Topology Seminar, EPFL, Lausanne, Switzerland. Talk: Sheaves as Computable and Stable Desciptors of Data.
  • June 2020 AATRN Seminar, online. Talk: Sheaves as Computable and Stable Desciptors of Data, video available here.
  • June 2020 Representation Theory Seminar, Bielefeld University, Bielefeld, Germany. Talk: Derived Methods for Persistence.
  • January 2019 Applied Topology workshop, Kyoto University, Kyoto, Japan. Talk: A Derived Isometry Theorem for Constructible Sheaves Over R.
  • May 2018 Bridging Statistics and Sheaves workshop, IMA, Minneapolis, USA. Poster: A Derived Isometry Theorem for Constructible Sheaves Over R.
  • March 2018 Luxembourg University, Luxembourg. Gave a 10 hours course with Steve Oudot: Theoretical Foundations of Persistence.
  • February 2018 Linking Topology to Algebraic Geometry and Statistics workshop, MPI, Leipzig, Germany. Poster: Computing the convolution distance for constructible sheaves over R.
  • December 2017 Journées de Géométrie Algorithmique, Aussois, France. Talk: Stable resolutions of multi-parameter persistence modules.
  • August 2017 Developing abstract foundations for TDA, Banff Center, Canada. Talk: Stable resolutions of multi-parameter persistence modules , video available here.
  • March 2017 Persistent homology working group, IHP, Paris. Talk: Stable resolutions of multi-parameter persistence modules.