学术会议
【2024.06-12 北京】Mathematical Theory of Deep Learning
发布时间:2024-06-06

This half-year program will focus on the basic mathematical theories and methods that have the potential to bridge the significant gap between the relatively limited theoretical foundation and the revolutionary practical success of Deep Learning (DL) and Large Language Models (LLMs). Topics of the program include, but are not limited to

lExpressive Power of Deep Neural Networks and LLMs, including approximation to continuous functions, memorization of finite dataset, capacity measures like VC dimension and Rademacher complexity.

lStatistical Property for DL and LLMs, generalization bounds in various scenarios: including algorithmic-free or algorithmic-dependent, i.i.d. dataset or OOD dataset, and overparametrized models.

lOptimization Methods for Training, including convergence rate, consistency of the learning process, computational complexity, and training for overparametrized models.

lTheory of LLLs, including the scaling law and the 'emergency' phenomena.

lDL with Rich Mathematical Structures, including graph neural networks, geometric learning, and PDE-informed learning

lRobustness and Security of DL and LLMs, including theory of adversarial learning, provable robustness, trade-off between accuracy and robustness.

lAI-driven Computation and Reasoning, including DL-driven PDE solving, optimization, optimal control, and deep reasoning.

Organization Committee:

Advisors:

Zhi-Ming Ma, Ruqian Lu, Lei Guo, Ya-Xiang Yuan

Organizers:

Xiao-Shan Gao, Yangyang Li, Xin Liu, Zhi-Xin Liu, Qi Meng, Manolis C.Tsakiris,

Shihua Zhang, Xinyu Zhang, Lihong Zhi, Chuan Zhou, Tao Zhou

Contact:

       Jing-An Li (010-82541731, ncmis@amss.ac.cn)

Activities:

The program consists of a flagship conference: “International Conference on Mathematical Theory of Deep Learning” and several short programs on specific topics.

The program is organized by

Hua Loo-Keng Center for Mathematical Sciences

Academy of Mathematics and Systems Science, Chinese Academy of Sciences https://amss.cas.cn/



附件下载: