报告题目:Error-free Training for Artificial Neural Networks

报告人：邓波 教授（内布拉斯加大学林肯分校）

报告时间：2024年6月26日  15:30-16:30

报告地点：数学学院205

报告摘要:  Models of Artificial Neural Networks play an essential role in Artificial Intelligence.

All ANN models must be trained before they are deployed to perform tasks. The majority of AI training is supervised. For large-scale models, there are no known methods to achieve 100% accuracy for supervised training. In this talk, I will discuss a newly discovered method that can train ANN models to perfect precision. I will outline the ideas from Dynamical Systems that guarantee the convergence of the error-free training algorithm, and show simulations on the most popular benchmark data for training algorithms in the field. I will also discuss the relationship between the ANN training problem and the classification problem of finite points in Euclidean space that is based on the Stone–Weierstrass approximation theorem in Analysis.

专家介绍：邓波，博士，内布拉斯加大学林肯分校教授。1977年应届毕业生考进复旦数学系，1981年考取教育部留美研究生，1987年拿到MSU博士学位，师从周修义教授。1987年-1988年在布郎大学做博士后，合作导师Jack Hale。研究方领域是动力系统、生物数学、人工智能。

【编辑：可视化与虚拟现实四川省重点实验室】