Barcelona Computational, Cognitive and Systems Neuroscience (BARCCSYN) 2025

To make a decision we must often combine diverse factors such as sensory inputs, past actions, and estimates of value. There is increasing evidence that the brain does this via a simple operation: give each factor a weight, sum the results, use the sum to set the log odds of a coin, and flip the coin. This operation is called logistic classification. It is common in machine learning and economics, and has close cousins in psychology. I will illustrate this computation in multiple species, and will describe it in more detail in mice that perform an audiovisual decision task. Through large-scale recordings and localized inactivations, I will show how the log odds of a choice are progressively computed at key stages in visual cortex, auditory cortex, prefrontal cortex, and superior colliculus. The results point to a single view of how the brain makes a variety of decisions.

Every movement we make requires us to precisely coordinate muscle activity across our body in space and time. In this talk I will describe our efforts to understand how the brain generates flexible, coordinated movement in dynamic environments. We have taken a behavior-centric approach to this problem, starting with the development of quantitative frameworks for mouse locomotion (LocoMouse; Machado et al., eLife 2015, 2020) and locomotor learning, in which mice adapt their locomotor symmetry in response to environmental perturbations (Darmohray et al., Neuron 2019). Combining this behavior-first approach with neurophysiology and genetic circuit dissection has revealed surprising insights into how cerebellar circuits flexibly coordinate whole-body movements, enabling adaptive behavior in dynamic environments. 

Part of the secret of human intelligence is its systematicity and productivity, that is, our ability to assemble new ideas from existing building blocks. Given that neural computation happens in largely unstructured neural networks, it remains a grand challenge to understand how the human brain can ‘compose’ new knowledge. In this talk, I will discuss experiments in which we tackle this question at the neural and computational levels. I will show how specific training regimes (curricula) facilitate composition, and provide evidence two mechanisms of compositional generalisation, a low-dimensional route that relies on cortex and a high-dimensional one that relies on hippocampus.