Computational Neuroscience Group

Overview

The Computational Neuroscience Group develops quantitative theories of neural computation at multiple scales: from the biophysics of ion channels and dendritic integration, through the dynamics of recurrent microcircuits, to the large-scale network mechanisms underlying perception, memory, and executive control. The group is equally at home with analytical methods from dynamical systems theory, statistical mechanics, and information theory as it is with machine-learning-inspired models and direct data fitting to electrophysiological recordings.

A defining feature of the group's programme is the close interplay between theory and data. Collaborators at partner neuroscience institutes supply spike-train recordings, calcium imaging, and EEG datasets; the group's models are constrained by and tested against this empirical evidence. Where models succeed, they generate concrete, experimentally testable predictions. This reciprocal workflow has produced a series of experimentally validated theories of working memory, attention gating, and sensory adaptation.

The group works in close collaboration with the Perception & Decision Lab and the Neural Engineering Group within the Cognitive & Neural Science division, and draws heavily on the Meridian HPC Cluster for large-scale simulations.

Research themes

• Recurrent neural network models of working memory and prefrontal dynamics
• Theory of predictive coding and hierarchical inference in cortical circuits
• Spike-timing-dependent plasticity rules and their role in unsupervised learning
• Low-dimensional manifold structure in population activity during motor planning
• Attractor network models of hippocampal memory consolidation
• Normative theories of perception under metabolic constraints (efficient coding)