CRCNS: DYNAMICS AND NEURAL MECHANISMS OF DECISION MAKING

描述（由申请人提供）：做出灵活决定的能力受损是许多精神障碍的特征，包括抑郁症、强迫症、自闭症和精神分裂症。然而，人们对负责理性决策的神经机制知之甚少。通过结合计算建模和单神经元记录灵长类动物行为的方法，该合作建议寻求获得关于大脑如何评估替代行动的预期结果并在面对高度动态和互动环境时做出最佳选择的新见解。大脑中的感觉和运动结构显示出许多优化设计系统的特征。同样，负责行动选择的大脑机制可能会采用最优计算策略，这种策略可以根据预期奖励进行动态调整。最近，与强化学习算法中某些元素相关的信号已经在大脑的各个区域被发现，例如脑干多巴胺神经元发出奖励预测错误的信号。拟议的研究计划将汇集正式模型（博弈论和强化学习），行为灵长类动物的电生理学，以及大规模皮质网络的基于生物物理学的计算建模。这些研究将(1)进一步发展灵长类动物基于竞争性游戏的决策任务范式；(2)检查额叶几个关键区域单个神经元的活动，以确定动态决策计算步骤的神经基础；(3)通过实施基于奖励的突触学习规则，开发基于生物物理的动态决策皮层网络模型。(4)研究导致选择行为随机性的可能机制，如不规则的尖峰活动和突触学习规则的随机性；(5)研究一个独特的神经网络被赋予跨试验时间整合能力的可能性，通过经验来估计预期奖励。

DESCRIPTION (provided by applicant): Impaired abilities to make flexible decisions characterize many mental disorders, including depression, obsessive-compulsive disorders, autism, and schizophrenia. Nevertheless, the neural mechanisms responsible for rational decision-making are poorly understood. By combining methods of computational modeling and single-neuron recording from behaving primates, this collaborative proposal seeks to obtain novel insights as to how the brain evaluates the expected outcomes of alternative actions and make optimal choices in the face of a highly dynamic and interactive environment. Sensory and motor structures in the brain display many features of optimally designed systems. Similarly, the brain mechanisms responsible for action selection might adopt optimal computational strategies that can be dynamically adjusted based on expected rewards. Recently, signals related to some elements in reinforcement learning algorithms have been identified in various brain areas, such as brainstem dopamine neurons signaling reward prediction errors. The proposed research program will bring together formal models (game theory and reinforcement learning), electrophysiology in behaving primates, and biophysically-based computational modeling of large- scale cortical networks. The proposed studies will (1) further develop the primate paradigm of a decision- making task based on competitive games, (2) examine the activity of single neurons in several key areas in the frontal lobe to identify neural basis of computational steps in dynamic decision-making, (3) develop a biophysically-based cortical network model for dynamic decision-making by implementing reward-based synaptic learning rules, (4) examine possible mechanisms responsible for the randomness of choice behavior, such as irregular spike activity and stochasticity in the synaptic learning rules, and (5) investigate the possibility that a distinct neural network is endowed with the ability of cross-trial temporal integration to estimate expected reward through experience.