17/05/24 Seminar by Dan Goodman

Friday 17th May 2024 at 11 AM, INT Henri Gastaut meeting room

Dan Goodman (Imperial College, London)

(invited by Laurent Perrinet) 

From neural mechanism to function

Abstract: What is the role of theoretical or computational work in neuroscience? In this talk, I will discuss how I think that techniques from machine learning can let us relate neural mechanisms to functions, illustrated with some examples from our recent research. In brief, theory lets us answer “what if” questions that would have been impossible to answer using experiments alone. Machine learning specifically gets us close to being able to answer “how well could the brain function if” questions that allow us to get at the computational role of particular neural mechanisms. Critically, this approach to theory is different to the ones that have often been suggested for computational neuroscience: fitting data, predicting data, letting us carry out “in silico” experiments. In summary, I will argue that if you want to understand what the functional role of a neural mechanism is, computational theory is essential, and that thinking of theory in this way lets us do it better. The research I will discuss includes: (1) How the heterogeneity of neurons allows the brain to solve sensory tasks with temporally rich structure. (2) How the structural modularity of the brain is not sufficient to guarantee functional modularity. (3) How neural nonlinearities are critical to extract temporally structured multimodal information.