From Variational Inequality Problem to Machine Learning Problem
演讲者
Jeremiah Nkwegu Ezeora
时间
2026年06月22日 10:00 至 11:00
地点
A3-4-101
线上
Zoom 204 323 0165
(BIMSA)
摘要
Fixed point problems and variational inequalities are cornerstones of nonlinear analysis, while machine learning optimization drives modern AI, their deep connections are often downplayed. This talk aims to unveil the profound mathematical equivalence between Fixed Point Problems, Variational Inequalities, and Machine Learning optimization. Through classical derivations and concrete algorithmic examples, including Gradient Descent, and Proximal Gradient methods, we demonstrate that training modern AI models is fundamentally a fixed-point iteration solving a variational inequality problem. This unified perspective not only clarifies the theoretical advantages of popular algorithms but also provides a good way of designing novel optimization methods(AI-models) with guaranteed convergence and stability.