Representation and propagation of uncertainty information in cortical populations

 描述（由申请人提供）：理解大脑用于表示和计算不确定信息的机制标志着系统神经科学的基本探索。生物体经常根据观察结果做出决定，这些观察结果由于嘈杂的传感器和世界上的模糊性而具有内在的不确定性。为了最佳地执行这些任务，大脑有必要代表和利用感官不确定性的知识。行为研究表明，在某些任务中，人类的表现接近最佳状态，这意味着大脑在一次又一次的尝试中表现和利用不确定性。概率群体编码（PPC）的理论框架假设大脑通过对刺激表示“似然函数”来编码群体活动模式中的感觉信息。虽然PPC已被用于构建使用神经似然操作的几种贝叶斯计算的实现，但没有人口水平的生理证据表明这是大脑使用的编码方案。 该项目结合了电生理学（多神经元记录）和计算神经科学，旨在阐明大脑如何在视觉决策过程中表示和计算感官不确定性。该项目将使用我们实验室以前设计的简单方向分类任务。这项任务的最佳表现要求观察者在逐个试验的基础上利用感官的不确定性，人类和猴子已经被证明可以接近最佳表现。在目标1中，初级视觉皮层中的群体记录将与行为测量相结合，以确定感觉不确定性是否在该区域中编码，具体而言，根据PPC作为似然函数。目的二是利用前额叶皮层的同步记录来检验V1和前额叶皮层之间是否存在由于共享不确定性信息编码而产生的功能相关。综上所述，拟议的工作将是迄今为止感知决策中贝叶斯计算的最全面的测试，特别关注PPC作为要测试的领先编码框架。了解大脑如何实现感觉的不确定性，有望迎来新一代的神经网络，并可能应用于开发神经假体设备，改善皮层人群的解码。

 DESCRIPTION (provided by applicant): Understanding the mechanisms used by the brain to represent, and compute with, uncertain information marks a fundamental quest in systems neuroscience. Organisms often make decisions based on observations that are inherently uncertain due to noisy sensors and ambiguity in the world. To optimally perform such tasks, it is necessary for the brain to represent and utilize knowledge of sensory uncertainty. Behavioral studies have demonstrated that in certain tasks, humans perform close to optimally, implying that the brain represents and utilize uncertainty on a trial-by-trial basis. The theoretical framework of probabilistic population coding (PPC) postulates that the brain encodes sensory information in the pattern of population activity by representing a "likelihood function" over the stimulus. Although PPCs have been used to construct implementations of several Bayesian computations using neurally plausible operations, there has been no population-level physiological evidence that this is the coding scheme used by the brain. The proposed project combines electrophysiology (multi-neuronal recordings) and computational neuroscience with the goal to elucidate how the brain represents and computes with sensory uncertainty during visual decision-making. The project will use a simple orientation classification task previously designed in our laboratories. Optimal performance on this task requires the observer to utilize sensory uncertainty on trial-by- trial basis, and human and monkeys have been shown to perform near optimally. In Aim 1, population recordings in primary visual cortex will be combined with behavioral measurements to determine whether sensory uncertainty is encoded in this area, specifically, as likelihood functions in accordance to PPC. In Aim 2, simultaneous recordings from prefrontal cortex will be used to test whether there exists functional correlation between V1 and prefrontal cortex due to shared uncertainty information encoding. Taken together, the proposed work will be the most comprehensive test to date of Bayesian computation in perceptual decision- making, with a particular focus on PPC as the leading encoding framework to be tested. Understanding how the brain implements sensory uncertainty promises to usher in a new generation of neural networks, with possible applications to developing neuroprosthetic devices with improved decoding from cortical populations.