当前位置 :

新闻中心

关于中科院神经所研究员杨天明做学术报告的通知

  讲座题目:Modeling Orbitofrontal Cortex and Model-based Reinforcement Learning with Reservoir Network

  主讲嘉宾:杨天明(中科院神经所研究员)

  时间:2016年11月7日(周一)10:00-11:00

  地点:科大西区科技实验西楼一楼117会议室

  报告摘要:

  Recently, model-based reinforcement learning (mbRL) has been used successfully in explaining animals’ learning behavior. In mbRL, the model, or the structure of the task, is used to evaluate the associations between actions and outcomes. However, in most studies, it is assumed that the task model is known. It is not well understood how animals acquire the model information necessary to build the reinforcement learning frame work in the first place, or how the model information is represented in the brain. Here, we propose a neural network model based on reservoir networks to solve this problem. Reservoir networks contain randomly and sparsely connected excitatory and inhibitory neurons. The heterogeneous and dynamic response patterns of neurons in a reservoir network makes it a perfect candidate to encode task information. Its linear readout is trained by a reinforcement learning algorithm to determine actions. Here, we demonstrate how a reservoir network that receives a reward input in addition to sensory inputs may achieve model-based reinforcement learning, and how the task model information is stored in the weights of its readout. In addition, we show that the reservoir network resembles features of orbitofrontal cortex (OFC). In particular, it explains how the OFC may encode task state space as proposed recently. Therefore, we propose our reservoir network model as a valid model of the OFC.

  嘉宾简介:

  杨天明博士于1992年至1997年间就读于复旦大学生物化学系,并获理学学士学位。2003年于美国贝勒医学院获得神经科学博士学位。2003-2008年期间,在美国西雅图华盛顿大学从事博士后工作。2008年底受聘于美国国家心理卫生研究所神经心理学研究室,任科学家一职。2013年起全职返回中国,就任中国科学院上海生命科学研究院神经科学研究所研究员。杨天明博士的研究兴趣主要在于抉择的神经机理。