Neural Representation of Uncertainty in Mouse Visual Cortex

当我们与环境互动时，我们需要评估我们决策的潜在后果。然而，我们的决定通常是基于不全面的信息，并且可能以不同的方式含糊不清。例如，当我们在公寓里寻找钥匙时，我们必须考虑到钥匙可能部分或完全被遮住了，我们可能对钥匙放在哪里，或者最后一次丢失时在哪里找到的记忆不完善。各种形式(知觉、运动、认知)的这种不确定性是大脑必须解决的根本挑战，以确保行为保持适应环境的限制和要求。在这个跨学科的项目中，我们建议将神经生理学家(Rochefort Lab)和计算神经学家(Lengyel Lab)的专业知识结合起来，以解决神经科学中的一个关键问题：大脑如何编码不确定性？大量文献支持这样的概念，即人类和其他动物在广泛的认知领域(从感觉感知到运动规划和执行)根据贝叶斯决策规则表示和传播不确定性(至少是近似的)做出决策。然而，这些计算背后的实际神经元表示仍然未知。特别是，尚不清楚大脑中的不确定性是机会主义的，即仅关于决策变量(决策过程最后阶段的变量)，还是普遍存在于信息处理的不同阶段，包括知觉不确定性、记忆不确定性以及与决策相关的变量。在人工智能中，这两种策略都有例子，有互补的优点和缺点。在解决大脑中的概率表征是机会主义的还是无处不在的问题上，一个限制进展的根本因素是缺乏能够区分决策不同阶段的不确定性的实验范式(例如，关于感知变量和决策变量的不确定性)。因此，该项目的主要目标是分析在一个专门为解决这一挑战而设计的范式中记录的行为和神经数据。我们以小鼠初级视觉皮质(V1)作为模型系统，在感知决策的背景下解决这个问题。我们将使用罗什福尔实验室获得的一个数据集，在该数据集中，当小鼠执行视觉引导的目标导向行为任务时，大量神经元群体的活动被记录在V1中。在对动物进行任务训练后，我们通过修改视觉刺激属性来操纵动物的感觉不确定性。该项目围绕两个目标组织：-目标1.分析行为数据，以获得感知和决策不确定性的逐次试验测量。-目标2.在小鼠V1中确定感知不确定性与决策不确定性的神经元表示。该项目将在神经生理学(Rochefort)和计算神经科学(Lengyel)实验室之间建立强有力的跨学科伙伴关系。这种合作将推动研究在细胞、网络和行为水平上全面了解成年人大脑中知觉不确定性的编码。它还将导致开发新的分析工具和新的计算模型，以实现大脑中决策的神经元执行。通过整合实验实验室和计算实验室的技能和专业知识，这两个实验室的成员将开发出这一研究领域高需求的新的跨学科专业知识。

When interacting with our environment, we need to evaluate the potential consequences of our decisions. However, our decisions are usually based on information that is partial and may be ambiguous in different ways. For example, when we are looking for our keys in the flat, we have to take into account that the keys may be partially or fully occluded, we may have imperfect memory about where we put them, or where we found them last time we lost them. Such uncertainty in various modalities (perceptual, motor, cognitive) presents a fundamental challenge that the brain must tackle in order to ensure that behaviour remains adapted to the constraints and demands of the environment. In this interdisciplinary project, we propose to bring together the expertise of neurophysiologists (Rochefort lab) and computational neuroscientists (Lengyel lab) to tackle this critical question in neuroscience: how does the brain encode uncertainty?A rich body of literature supports the notion that humans and other animals make decisions based on representing and propagating uncertainty (at least approximately) according to the rules of Bayesian decision making, in a broad range of cognitive domains (from sensory perception to motor planning and execution). However, the actual neuronal representations underlying these computations remain unknown. In particular, it is unclear whether uncertainty in the brain is represented opportunistically, i.e. only about decision variables (variables at the final stages of the decision-making process), or whether it is represented ubiquitously, throughout different stages of information processing, including perceptual uncertainty, memory uncertainty, as well as over variables relevant to the decision. In artificial intelligence, there are examples for both strategies, with complementary advantages and disadvantages.A fundamental factor limiting progress in resolving whether probabilistic representations in the brain are opportunistic or ubiquitous is the lack of experimental paradigms that can distinguish between uncertainty at different stages of decision making (e.g. uncertainty about perceptual vs. decision variables). Thus, the main goal of the project is to analyse behavioural and neural data recorded during a paradigm that has been specifically designed to address this challenge.We address this question in the context of perceptual decision making, using the mouse primary visual cortex (V1) as a model system. We will use a dataset obtained in the Rochefort lab, in which the activity of large neuronal populations was recorded in V1 while mice were performing a visually-guided goal-directed behavioural task. After training animals on the task, we manipulated the sensory uncertainty of the animals by modifying visual stimulus properties. The project is organised around 2 aims:- Aim 1. Analyse behavioural data in order to obtain trial-by-trial measures of perceptual and decision uncertainty.- Aim 2. Determine the neuronal representation of perceptual vs. decision uncertainty in mouse V1.This project will enable a strong interdisciplinary partnership between a neurophysiology (Rochefort) and a computational neuroscience (Lengyel) lab. This collaboration will drive research to comprehensively understand the encoding of perceptual uncertainty in the adult brain at the cellular, network, and behavioural level. It will also lead to the development of new analysis tools and new computational models of the neuronal implementation of decision-making in the brain. By integrating the skills and expertise from an experimental and a computational lab, members of both labs will develop new interdisciplinary expertise that is of high demand in this field of research.