Report time:July 8, 10:30 am
Reporting location: Computer BuildingConference Room 313
Report title: Deep bet365 login games with universal approximation properties: first-order optimization method
pick To:
bet365 login games is the cornerstone of the success of deep neural networks。However，Almost all existing deep neural network design methods ignore the property of universal approximation。We propose a unified framework，Based on the first-order optimization algorithm to design a deep neural network architecture with guaranteed bet365 login games。What we get is that deep neural networks are all width-bounded，That is, its width will not increase with the improvement of approximation accuracy，So it is close to the current common actual scenarios。In addition，Add normalization to the network、Operations such as downsampling and upsampling do not damage the bet365 login games。As far as we know，This is the first work to design width bounded networks with universal approximation guarantee in a principled manner。Our framework can inspire various neural network architectures，includes famous onesResNet and DenseNet, etc.
About the speaker:
Lin Zhouchenis the deputy dean of the School of Intelligence, Peking bet365 login games，Boya Distinguished Professor，Research fields are machine learning and computer vision。He has published papers in core artificial intelligence journals and conferencesMore than 300 articles，Published 5 monographs in Chinese and English，Google citation number is 34,More than 000 times。He has served as the area chair and senior area chair for many top conferences in the industry。He won the first prize in natural science in the 2023 CAAI and 2020 CCF Science and Technology Awards。He is the director of the Machine Vision Committee bet365 login games Society of Image and Graphics (CSIG)，Deputy Director of the Pattern Recognition and Machine Intelligence Committee bet365 login games Society of Automation，IAPR、IEEE、fellow of CSIG and AAIA，National Outstanding Youth，Responsible for the Ministry of Science and Technology’s Science and Technology Innovation 2030-“New Generation Artificial Intelligence” major projectPeople.
