Eva Dyer
Title- Representation learning and alignment in biological and artificial neural networks
Abstract: In both biological and artificial neural networks, we are faced with similar challenges in interpreting how the representations of many neurons (or units) change across different domains, across perturbations, or across different individuals (or networks) performing the same task. The question, however, of how we should go about comparing the activities of populations of neurons over all these differing conditions is still a major challenge. A critical observation is that when the activity of many neurons can be modeled as being driven by a smaller number of latent factors, then distinct measurements acquired from neurons that share the similar underlying latent space can be compared by finding and aligning their latent factors. In this talk, I will highlight new approaches that my lab is developing for representation learning and alignment, and demonstrate their applications in the analysis and interpretation of neural networks. Being able to align neural representations promises meaningful ways of comparing high-dimensional neural activities across times, subsets of neurons, or individuals.