题    目：Approximation Ability of Transformer Networks

主讲人：古林燕 副教授

单    位：中山大学

时    间：2025年9月26日 17:00

地    点：7号楼7303

摘    要：In this talk, we explore the approximation capabilities of Transformer networks. Specifically, we construct Transformer architectures to approximate maxout neural networks. Through the relationship between these two types of networks, we further investigate how well Transformers approximate piecewise linear functions and analyze the impact of network complexity on the number of linear regions.

简    介：古林燕，中山大学理学博士，中山大学副教授，硕士生导师。主要从事深度学习的理论和应用、神经网络逼近理论的研究，主持国家自然科学基金项目1项，广东省自然科学基金2项，在IEEE Transactions on Neural Networks and Learning Systems、Neural Networks、Physics of Fluids等期刊发表论文10余篇。