【学术视频】第六届复杂系统统计物理与数学国际研讨会 | 香港科技大学王國彝教授

原创 KouShare 蔻享学术

更多精彩视频登陆网站

www.koushare.com

图 | K. Y. Michael Wong

题目：Multisensory integration and predictive information in neural systems

报告人：K. Y. Michael Wong（王國彝）

单位：Hong Kong University of Science and Technology

时间：2020-01-15

地点：华侨大学

扫码观看精彩报告视频

报告摘要

Neural systems gather information from different channels resulting in enhanced reliability. The optimal estimate is given by Bayes' rule, and remarkably, experiments on macaques showed that the brain can achieve this optimum. It is therefore interesting to consider the neural architecture and mechanism underlying this feat. I will discuss the results of our study on a decentralized network architecture. In the VIP and MSTd areas of the brain, experiments showed that both congruent and opposite neurons exist. The preferred responses of congruent neurons to visual and vestibular cues are the same, and contribute to multisensory integration. The preferred responses of opposite neurons to the visual and vestibular cues are opposite and do not contribute to multisensory integration. Yet they are equally populous as the congruent neurons. I will introduce a model to explain the role played by opposite neurons in segregating disparate signals. In the second part of the talk, I will discuss predictive information in the retina. Experiments showed that the retina of salamander and rabbit not only receive visual signals, but also process the information they received by making predictions. We performed experiments on the retina of bullfrog and found that this 'anticipative coding' phenomenon depends on the dynamics of the input. When the bullfrog retina is shown a moving bar, whose positions are calculated from the Hidden Markov Model (HMM), the mutual information between the responses of the bullfrog retina and the input positions as a function of their time differences reveal that the responses of the retina have correlations with subsequent visual signals. However, when the moving bar positions are calculated from Ornstein-Uhlenbeck (OU) process, the correlations disappear. We propose a neural network model to simulate this kind of anticipative behaviors.

个人简介

Professor Michael Wong obtained his BSc degree from the University of Hong Kong and his MSc and PhD degrees from the University of California, Los Angeles. He did his post-doctoral research in Imperial College London and the University of Oxford. He joined the Department of Physics of HKUST in 1992. In 2005, he received the School of Science Teaching Award of HKUST. His research areas include physics of complex and disordered systems; statistical physics; computational neuroscience; optimization; spin glasses; optimal control of telecommunication networks, transportation networks and power grids.