Conference on the Mathematical Theory of Deep Neural Networks

The Conference on the Mathematical Theory of Deep Neural Networks brings together mathematicians, statisticians, and theoretical machine learning researchers to develop rigorous foundations for understanding why deep learning works and what its fundamental limits are. Topics include approximation theory for neural networks, optimization landscapes, generalization bounds, implicit regularization, expressivity of network architectures, connections to statistical mechanics, and the geometry of learned representations. The conference responds to a widely recognized need for mathematical clarity in a field that has advanced largely through empirical discovery.

This event is ideal for researchers with strong mathematical backgrounds who want to engage seriously with deep learning theory, as well as deep learning practitioners who want to understand the theoretical principles underlying their tools. Graduate students in mathematics, statistics, and computer science will find the conference an excellent venue for exposure to open problems and for establishing collaborations across disciplinary boundaries. The work presented here has lasting value for the long-term scientific understanding of artificial intelligence.