Non-Linear and High Dimensional Inference

Non-Linear and High Dimensional Inference

3-7 October 2022 - IHP, Paris

In statistical learning and modern statistics, it is usually assumed that although data is observed in very high dimension, it actually live on — or near — a geometric structure of low dimension. The knowledge of such a low-complexity structure is of fundamental importance to improve both theoretical and computational aspects of a learning procedures.

Program 

• Abstracts
• General Info handout

Invited Speakers

• Peter Bartlett (UC Berkeley)
  • The Dynamics of Sharpness-Aware Minimization (video, slides)
• Mikhail Belkin (UC San Diego)
  • Neural networks, wide and deep, singular kernels and Bayes optimality (video, slides)
• Denis Belometsny (Duisburg-Essen)
  • Dimensionality reduction in reinforcement learning by randomisation (video)
• Clément Bérenfeld (CEREMADE, Dauphine) 
  • Understanding the geometry of high-dimensional data through the reach (video, slides)
• Gilles Blanchard (LMO, Orsay)
  • Stein effect for estimating many vector means: a "blessing of dimensionality" phenomenon (video, slides)
• Alexander Cloninger (UC San Diego) 
  • Learning on and near Low-Dimensional Subsets of the Wasserstein Manifold (video, slides)
• Adeline Fermanian (Mines ParisTech) 
  • Scaling ResNets in the Large-depth Regime (video, slides)
• Damien Garreau (U Côte d'Azur)
  • What does LIME really see in images? (video, slides)
• John Harlim (Penn State) 
  • Leveraging the RBF operator estimation for manifold learning (video, slides)
• Claire Lacour (Paris-Est, Marne-la-Vallée)
  • On the use of overfitting for estimator selection (video, slides)
• Sophie Langer (TU Darmstadt) 
  • Overcoming the curse of dimensionality with deep neural networks (video, slides)
• Marina Meila (University of Washington) 
  • Manifold Learning, Explanations and Eigenflows (part 1, part 2, slides)
• Franck Picard (ENS Lyon) 
  • A probabilistic Graph Coupling View of Dimension Reduction (video, slides)
• Wolfgang Polonik (UC Davis) 
  • Topologically penalized regression on manifolds (video)
• Judith Rousseau (Oxford) 
  • Bayesian nonparametric estimation of a density living near an unknown manifold (video)
• Johannes Schmidt-Hieber (Twente) 
  • A statistical analysis of an image classification problem (video, slides)
• Claudia Strauch (Aarhus) 
  • On high-dimensional Lévy-driven Ornstein–Uhlenbeck processes (slides)
• Botond Szabó (Bocconi) 
  • Linear methods for nonlinear inverse problems (video, slides)
• Vasiliki Velona (Hebrew University of Jerusalem) 
  • Learning a partial correlation graph using only a few covariance queries (video, slides)
• Nicolas Verzelen (Inrae Montpellier)
  • Optimal Permutation Estimation in Crowd-Sourcing problems (video, slides)

Free Afternoon & Lunch (Info handout)

• Institut Henri Poincaré’s neighborhood has plenty of diverse food places.
  The two-hour lunch breaks give you enough time to eat in a restaurant, or to grab something to take away.
• Wednesday afternoon is free! Feel free to stay at IHP to chat or to explore Paris.