2018年11月29日学术报告——Wen-Hao Zhang
题 目:Towards a Complete Theory of Multisensory Processing in the Brain
报告人:Wen-Hao Zhang
时 间:2018年11月29日(周四) 10:30
地 点:仙林校区计算机学科楼422房间
报告人简介:
Dr. Wen-Hao Zhang is a postdoc researcher in the Department of Mathematics at University of Pittsburgh. He received his PhD from Institute of Neuroscience, Chinese Academy of Sciences on 2015. And then he worked in Department of Physics at Hong Kong University of Science and Technology during 2015-2016. Then he worked in Carnegie Mellon University from 2016-2018. His current research interest includes nonlinear neural network dynamics, neural coding, Bayesian inference and its neural implementation.
报告内容:
Our brain perceives the world by exploiting multiple sensory modalities to extract information about various aspects of external stimuli. Extensive studies indicate that the information processing in the brain is implementing probabilistic inference: the network of neurons compute the posterior belief over unobserved causes given observations. In multisensory processing, if sensory inputs are from the same stimulus of interest, they should be integrated to improve perception; otherwise, they should be segregated to distinguish different stimuli. In reality, however, the brain faces the challenge of recognizing stimuli without knowing in advance whether sensory inputs come from the same or different stimuli. How multisensory processing is implemented in neural circuitry remains a challenge. In this talk, I would like to present my recent effort in pursuing this question through combining theory and experimental evidence. Experiments suggest that there are several multisensory brain areas that are simultaneously involved in multisensory integration. And within each brain area, there are two distinguishing groups of neurons, congruent neurons which exhibit integrative responses and opposite neurons whose functions remain mystery yet. Inspired by experiments, we propose a decentralized network composed of several inter-connected modules, with each module modelling a multisensory brain area. In our model, the multisensory integration is achieved by the cross-talk between congruent neurons across network modules, while opposite neurons compute the disparity between multisensory inputs. And eventually the concurrent integration and segregation emerge locally from the interplay between two types of neurons across network modules. Through this process, the brain achieves rapid stimulus perception if the inputs come from the same stimulus of interest, and differentiates and recognizes stimuli based on individual input with little time delay if the they come from different stimuli of interest. Our study may shed new light on a new architecture for multisensory processing in artificial intelligence.