Watch a 45-minute conference talk exploring classification problems in high-dimensional spaces through the lens of deep neural networks, presented at the Centre International de Rencontres Mathématiques. Delve into three fundamental aspects of classification: decision boundary complexity, noise, and margin conditions. Learn how classification problems can be efficiently approximated in high dimensions under specific decision boundary conditions, and discover how margin conditions enable rapid approximation rates despite high dimensionality and discontinuity. Examine the extension of approximation results to learning outcomes, with insights into optimal learning rates for empirical risk minimization in high-dimensional classification. Recorded during the "SIGMA, Signal, Image, Geometry, Modeling, Approximation" thematic meeting in Marseille, France, this mathematical presentation is available with chapter markers, keywords, abstracts, and comprehensive bibliographic references through CIRM's Audiovisual Mathematics Library.
High-Dimensional Classification with Deep Neural Networks: Decision Boundaries
Centre International de Rencontres Mathématiques via YouTube
Overview
Syllabus
Philipp Petersen: High-dimensional classification with deeep neural networks: decision boundaries
Taught by
Centre International de Rencontres Mathématiques
Related Courses
-
Build Decision Trees, SVMs, and Artificial Neural Networks
-
Introduction to Neural Networks
-
High-Dimensional Learning and Deep Neural Networks
-
Phase Diagram of Stochastic Gradient Descent in High-Dimensional Two-Layer Neural Networks
-
Neural Networks as Interacting Particle Systems
-
Neural Set Function Extensions: Learning with Discrete Functions in High Dimensions
