【题目】Deep Learning Beyond the Scaling Law: A Neuro-inspired Solution

【摘要】Although the invention of the neural network was to mimic a human's brain,the current development of deep learning is not primarily driven by the increasingly growing understanding to brain. Brain is the most intelligent system we have ever known so far,although the brain remains vastly undiscovered, it is clear that the existing deep learning still goes far behind human brain in many important aspects such as efficiency, interpretability, memory, etc. Given the incredible capability of the human brain, we argue that neuroscience can always offer support for deep learning as a think tank and a validation means. Clearly, the characters of the current mainstream deep learning models are fundamentally different from the biological neural system. One remarkable distinction is that the deep learning models lack the neuronal diversity that is everywhere in the human brain. Different from artificial networks that are built on a single universal primitive neuron type,the human brain has numerous morphologically and functionally diverse neurons. The neuronal diversity is an enabling factor for all kinds of intelligent behaviors. In this talk,I will discuss what values can the neuronal diversity potentially add to the artificial neural network,with an emphasis on the scaling law.

报告人简介

Dr. Fenglei Fan is currently an Assistant Professor in the Department of Data Science at the City University of Hong Kong. His research focuses on NeuroAI and its applications in data science. He obtained his Ph.D. from Rensselaer Polytechnic Institute in the United States in 2021. Dr. Fenglei Fan was supported by the IBM AI Horizon Fellowship during Ph.D.,and was invited to intern at MIT-IBM AI Watson Lab. He then conducted a one-year postdoctoral research at Cornell University. His main research outcomes have been published in flagship venues in the field of artificial intelligence and data science,such as JMLR,CVPR,IEEE TNNLS,IEEE TMI,IEEE TCI,and IEEE TAI. His doctoral dissertation won the 2021 Outstanding Doctoral Dissertation Award by the International Neural Network Society (INNS). His representative work was selected as 2024 CVPR Best Paper Award Candidate (26 out of 1w+ submissions),and he also won IEEE TRPMS Best Paper Award by the IEEE Nuclear and Plasma Society. Dr. Fenglei Fan has organized a tutorial at the top conferenceAAAI2023 andWWW2025. He also served as the senior PC member ofIJCAI2025.

【会议时间】2025/03/06 14:30-16:30 (GMT+08:00) 中国标准时间 - 北京

【点击链接入会】https://meeting.tencent.com/dm/qxsh01A8Dz4I

【腾讯会议】634-376-989

【主持人】刘宝东