Noise in neural codes: consequences for memory, exploration and decision-making.

Working memory (WM) refers to the nervous system's capacity to internally maintain information about the external world over short periods, and manipulate that information in support of behaviour. WM is considered fundamental to many if not all cognitive functions, and its impairment by age or disease has far-reaching consequences for human health and well-being. This research programme aims to understand how a fallible WM system arises from noisy neural coding, and how WM errors affect human ab ility to explore the environment and make decisions. The key goals are: To develop a detailed computational model of storage and retrieval from WM, based on empirical results obtained with elementary visual features. The aim is to go beyond a simple model of errors, to provide a neural account of binding failures, decay, time to decision, and the confidence assigned to memories. To investigate the impact of WM errors on value-based decision making. Here the internal representation is of the co sts and rewards associated with different possible courses of action. Using gambling tasks, we will investigate the nature of errors in representation of value, how these errors affect decisions, and what compensatory mechanisms mitigate their impact. To understand the most frequent decision made by the nervous system: where to direct gaze. Specifically, we will address how eye movements are selected during visual search, a microcosm of human choice behaviour that draws on WM both for visual pa rameters (where have I looked?) and for reward (where should I look next?).