Organisation des séminaires et calendrier

Speaker: Rémi Monasson (ENS, Paris)

Abstract: Despite the unsustainable growth in energy consumption by artificial intelligence models and the recognition of the major role played by metabolic constraints in brain evolution, the relationship between computation and energy remains insufficiently studied and understood. In this presentation, I will show two recent experiments that shed light on this relationship. More specifically, I will show how models of these circuits, inspired from statistical physics, can predict energy consumption under normal conditions and help explain resilience in the event of prolonged resource scarcity.

We revisit the calculation of spectral densities and heavy-heavy-light (HHL)
operator product expansion (OPE) coefficients in three-dimensional conformal field theories
using thermal one-point functions on S^1 x S^2. We develop systematic expansions in the regime of large ∆_H, and we validate our analytic tools by comparing the results with free field theories.

Beyond Hopfield networks for the modeling of neural computations

by Alexis DUBREUIL (CNRS, Institute for Neurodegenerative Diseases, Bordeaux)

Attractor neural networks, such as the Hopfield model, have been proposed to underlie various computations such as pattern recognition or working memory maintenance. Here I will introduce an extension of the Hopfiled model and show how it can be used to generically model computations in neural networks. I will discuss how it may shed light on the structure of brain circuits.

TBA

TBA

TBA

TBA