Computational studies of decision-making and planning in the primate brain

大脑是如何对行为做出决定的？一种假说认为，当我们看世界时，大脑的一部分会根据当前可能的行为来处理感官信息。这些行为的神经表征相互竞争，直到其中一个被选中来指导公开的表现。换句话说，行为是世界呈现给我们的需求和行动机会之间的一场持续的战斗。在我的实验室里，我通过两种方法来研究这个广泛的假设。一方面，我利用多电极记录、功能成像和行为分析进行实验。另一方面，我开发了计算模型来描述神经系统执行决策和计划的机制。这些模型通过数学方程来表达，这些方程被实现为模拟广泛的神经生理和行为现象的计算机程序。利用这些模型，我们可以尝试在机械层面上理解大脑的运作，甚至预测它在各种新情况下的行为。在这个NSERC申请中，我请求资金来支持我的计算建模研究项目。最近，我开发了一个数学模型来解释灵长类动物决策和计划的各种神经和行为结果。该模型为计算神经科学的长期研究计划提供了起点。它将涉及一系列的项目，包括：探索在不同表征参照系（例如从视觉坐标到运动坐标）之间学习感官信息转换的机制；通过强化学习扩展模型以适应；并赋予模型发现任意刺激-反应关联规则的能力。人们希望这些计算项目将演变成一个长期的研究项目，从而对自愿行为背后的大脑机制产生新的见解。

How does the brain make decisions about actions? One hypothesis suggests that when we look at the world, part of our brain processes the sensory information in terms of the actions which are currently possible. Neural representations of these actions then compete against each other until one is chosen to guide overt performance. In other words, behavior is a constant battle between the demands and opportunities for action which the world presents to us.      In my laboratory, I investigate this broad hypothesis through two approaches. On one hand, I conduct experiments utilizing multi-electrode recording, functional imaging, and behavioral analysis. On the other, I develop computational models which describe the mechanisms by which the nervous system implements decision-making and planning. These models are expressed through mathematical equations which are implemented as computer programs that simulate a wide range of neurophysiological and behavioral phenomena. Using these models, we can try to understand brain operation at a mechanistic level, and even to predict how it should behave in various novel situations.      In this NSERC application, I am requesting funds to support my computational modeling research program. Recently, I have developed a mathematical model which explains a variety of neural and behavioral results on primate decision-making and planning. This model provides the starting point for a long-term research program in computational neuroscience. It will involve a series of projects, including: exploring the mechanisms for learning the transformation of sensory information between different representational reference frames (e.g. from visual coordinates to motor coordinates); extending the model to adapt through reinforcement learning; and giving the model the capacity to discover arbitrary stimulus-response association rules. It is hoped that these computational projects will evolve into a long-term research program that will lead to new insights into the brain mechanisms underlying voluntary behavior.