题目(TITLE): Sequence Learning and Adaptive Decision Making: A Recurrent Neural Network Model 讲座人(SPEAKER): 杨天明 研究员，中国科学院神经科学研究所  主持人（CHAIR）: 余山 研究员，中国科学院自动化研究所  时间(TIME): 2019年11月5日，下午2:00-3:00  地点(VENUE): 自动化大厦13层第2会议室

Abstract

Our brain makes flexible and adaptive decisions in a complex environment. Because of this complexity, the current field lacks a unified model of how the brain achieves this function. In our current study,  we provide a novel modeling approach in which decision making is considered as a sequence learning problem. By training a neural network to predict future sensory, motor, and reward events,  we show that the network performs several tasks that have been well studied experimentally in the field, each of them covering a different aspect of decision making or learning.  The network model can account for both the behavior and neuron data in these tasks. We compare and discuss the results against a variety of modeling approaches in the field, including the drift diffusion model, attractor network model, Bayesian inference, and reinforcement learning. We show that our model, with one neural network architecture,  may account for experimental findings previously modeled with these diverse approaches and provide a unified framework for decision making and learning problems. 

Biography