Adaptation in decision circuits: temporal history and the efficiency of choice

 描述(由申请人提供)：在动态和变化的环境中，做出有效决策的能力至关重要，管理从简单到复杂的各种行为。此外，决策改变是许多疾病的标志，如癫痫、严重抑郁症和精神分裂症。特别是，价值导向的决策可以通过异常处理最近奖励的历史和利用过去的经验来指导未来决策的能力来改变。虽然新兴的研究已经勾勒出了参与决策的许多大脑区域，但神经回路如何决定还是个未知数。心理学、经济学和生态学的理论家概述了理性选择行为的标准模型，定义了最佳选择者应该如何行为以最大化结果。与这些理论预测相反，动物和人类的经验选择行为往往明显偏离最优。这种低效很可能反映了生物决策系统的限制，而且 研究违反理性的行为为了解决策的神经生物学基础提供了潜在的见解 制作。在这个应用中，我们检查以前的历史对决策过程及其基本电路的影响。我们假设，在编码价值的神经活动和选择行为中，时间语境依赖于神经回路代表价值的方式。具体地说，我们假设自适应价值编码是使用在感觉皮层回路中广泛存在的标准计算、分裂归一化来实现的，并且知觉加工中的适应为理解估值和决策中的适应提供了一个框架。为了验证这一假设，我们建议在神经、计算和行为水平上采取三个目标来解决价值编码中的适应问题。在目标1中，我们提出了电生理记录实验，以测试值的神经表示是否适应先前的奖励历史，以及这种计算是否与除法归一化算法相匹配。在目标2中，我们提出了计算模拟实验，以检验归一化模型在解释各种不同的值适应效应方面的通用性，并对自适应值编码对选择行为的影响做出具体预测。在目标3中，我们提出了在动物模型和人类受试者中的选择行为实验，以检验自适应值编码可以选择性地提高选择效率的预测。理解价值编码中的适应对于理解标准选择行为和病理性选择行为都至关重要。就理性选择理论而言，决策中的时间历史效应是次优的，但可能反映了一种更全局的最优化，它平衡了选择效率和运行生物决策过程的约束。这种时间依赖性可能对理解情感障碍特别重要，例如抑郁症和双相情感障碍，在这些疾病中，长期的低或高奖励状态可能会显著阻碍决策过程。

 DESCRIPTION (provided by applicant): The ability to make efficient decisions is critical in a dynamic and changing environment, governing behavior ranging from the simple to the complex. Furthermore, altered decision-making is a hallmark of a number of diseases, such as epilepsy, major depression, and schizophrenia. In particular, value-guided decisions can be altered by aberrant processing of the history of recent rewards and the ability to use past experience to guide future decisions. While emerging work has outlined many brain areas involved in decision-making, how neural circuits decide is unknown. Theorists in psychology, economics, and ecology have outlined standard models of rational choice behavior, defining how optimal choosers should behave to maximize outcomes. In contrast to these theoretical predictions, empirical choice behavior in animals and humans often deviates markedly from optimality. Such inefficiencies likely reflect the constraints of a biological decision system, and studying rationality violations offers potential insight into the neurobiological basis of decision making. In this application, we examine the effect of previous history on the decision process and its underlying circuits. We hypothesize that temporal context-dependence in both value-coding neural activity and choice behavior arises from the way neural circuits represent value. Specifically, we hypothesize that adaptive value- coding is implemented using a standard computation widely found in sensory cortical circuits, divisive normalization, and that adaptation in perceptual processing provides a framework for understanding adaptation in valuation and decision-making. To test this hypothesis, we propose to undertake three aims addressing adaptation in value coding at the neural, computational, and behavioral levels. In Aim 1, we propose electrophysiological recording experiments to test whether the neural representation of value adapts to prior reward history, and whether this computation matches the divisive normalization algorithm. In Aim 2, we propose computational modeling experiments which will test the generality of the normalization model in explaining various, different value adaptation effects, and make specific predictions about the effect of adaptive value coding on choice behavior. In Aim 3, we propose choice behavior experiments, in both an animal model and human subjects, to test the prediction that adaptive value coding can selectively enhance the efficiency of choice. Understanding adaptation in value coding is crucial for understanding both standard and pathological choice behavior. Temporal history effects in decision-making are suboptimal in terms of rational theories of choice, but may reflect a more global optimality that balances choice efficiency and the constraints of operating a biological decision process. Such temporal-dependence may be particularly important for understanding affective disorders, such as depression and bipolar disorder, where prolonged periods of low or high reward states may significantly impede the decision process.