Neural computations for working memory and decision making

• Main Author: Cavanagh, Sean Edward
• Format: Dissertation
• Language: English
• Published: UCL (University College London) 2019

It has long been known that neurons can possess spatial receptive fields, but whether they exhibit temporal receptive fields is relatively unexplored. Higher cognition relies upon processing across long timescales, such as: the holding of information in working memory, or the accumulation of evidence to form decisions. The temporal dynamics of individual neurons during these processes, and how they facilitate circuit-level computations, are pressing questions. This thesis will present four studies that try to address these questions using a variety of techniques: behavioural analyses, single neuron electrophysiology, pharmacology, and computational modelling. It will first present a method to index the temporal receptive field of single neurons, which highlights heterogeneity within prefrontal cortical regions. This heterogeneity is functionally significant, as neurons with longer timescales exhibit stronger and more sustained value correlates during choice. This may provide a neural mechanism for maintaining predictions and updating stored values during learning. A second study shows that the concept of temporal receptive fields can be used to reconcile competing accounts of neuronal activity during working memory tasks. A neuron's temporal receptive field is predictive of both its degree of task involvement, and its working memory coding dynamics. Next, a further behavioural study probed the role of visual attention in evidence accumulation - revealing fixations are drawn towards valuable and novel stimuli. Finally, data from a complex decision-making task is presented, where monkeys had to accumulate evidence across time. The function of N-methyl-D-aspartate (NMDA) receptors during this extended cognitive process was investigated using pharmacological manipulations. The induced behavioural changes were consistent with those predicted by a lowering of the excitation-inhibition (E/I) ratio in a spiking cortical circuit model. Together these results provide important insights into the role of neuronal timescales in high-level cognition, and have implications for cognitive deficits in neuropsychiatric disorders associated with cortical E/I dysfunction.