Mechanisms of learning a visual discrimination

描述（由申请人提供）：我们如何看待世界会受到我们的经历的强烈影响。我们可以学会将某些视觉图像与即将到来的奖励或惩罚联系起来（例如，看到糖果店的前面）。通过练习，我们还可以学会更好地区分微弱、嘈杂或模糊的图像（例如，一个有经验的观鸟者识别飞行中的鸟）。我们提出的研究目标是了解这两种看似不同的视觉学习形式的共同神经机制，一种是形成视觉输入和寻求奖励行为之间的联系，另一种是提高感知敏感性。关键的思想是，视觉信息是通过大脑中不同的神经元群来表示的，这两种形式的学习都涉及从这些神经元群中选择具有特定任务适当特征的神经元的过程。联想学习需要选择代表预测奖励的适当视觉特征的神经元。感知学习需要选择具有最高灵敏度的代表特定视觉特征的神经元。我们的三个具体目标将使用计算建模，人类心理物理学，并结合猴子的心理物理学和生理学来确定控制这一选择过程的神经机制。目标1将介绍一个计算模型，该模型可以通过基于引导行为最大化奖励的能力从感觉表征中选择输出来解释联想和感知学习。我们将以几种方式测试该模型，包括比较猴子学习视觉辨别任务的行为和神经数据，以及人类受试者学习类似任务的行为数据。该模型基于两个计算原理，这两个原理将指导其他两个目标中的实验。第一个原理是，学习源于感觉神经元和决策神经元之间功能连接的变化。目的2将测试两个皮质区域之间相互作用的变化是否反映了这些预测的功能连接变化。第二个原则是，学习是由一个识别预测和实际奖励之间差异的过程驱动的。目的3将测试皮层下结构（即尾状核）中的神经元是否在学习过程中编码这种奖励预测错误。这些研究将有助于统一先前不同的联想学习和知觉学习领域，并为允许经验塑造健康视觉系统功能的机制提供深远的视角。

DESCRIPTION (provided by applicant): How we see the world can be influenced strongly by our experiences. We can learn to associate certain visual images with impending rewards or punishments (e.g., seeing the front of a candy store). We can also learn to better differentiate weak, noisy, or obscure images with practice (e.g., an experienced bird- watcher identifying a bird in flight). The goal of our proposed research is to understand neural mechanisms common to these two seemingly different forms of visual learning, one that forms associations between visual input and reward-seeking behavior, the other that improves perceptual sensitivity. The key idea is that visual information is represented across heterogeneous populations of neurons in the brain, and both forms of learning involve a process of selecting neurons from these populations that have the appropriate characteristics for a given task. Associative learning requires selecting neurons that represent the appropriate visual features that predict reward. Perceptual learning requires selecting neurons that represent a particular visual feature with the highest sensitivity. Our three specific Aims will use computational modeling, human psychophysics, and combined psychophysics and physiology in monkeys to identify neural mechanisms that govern this selection process. Aim 1 will introduce a computational model that can account for both associative and perceptual learning by selecting outputs from a sensory representation based on their ability to guide behavior that maximizes reward. We will test the model in several ways, including a comparison to behavioral and neural data from monkeys learning a demanding visual discrimination task and behavioral data from human subjects learning a similar task. The model is based on two computational principles that will guide the experiments in the other two Aims. The first principle is that learning results from changes in functional connectivity between sensory and decision neurons. Aim 2 will test whether changes in interactions between two cortical areas reflect these predicted changes in functional connectivity. The second principle is that learning is driven by a process that identifies discrepancies between predicted and actual reward. Aim 3 will test whether neurons in a subcortical structure known as the caudate encode this kind of reward prediction error during learning. These studies will help to unify previously disparate fields of associative and perceptual learning and provide a far-reaching perspective on mechanisms that allow experiences to shape the functions of a healthy visual system. PUBLIC HEALTH RELEVANCE: The proposed work is basic research, designed to provide new insights into how a healthy nervous system learns from experience to more effectively process visual information. Thus, direct benefits to public health are intended to come in the longer term, as these new insights can be used to design new ways to diagnose and treat disorders of visual perception (i.e., visual agnosias) and learning.