Abstract

Please join us for our fortnightly journal club online via zoom where two presenters will jointly present a topic together. The next topic is ‘Non-normal dynamics in neural computations’ presented by Jake Stroud and Marine Schimel.

Zoom information: https://us02web.zoom.us/j/84958321096?pwd=dFpsYnpJYWVNeHlJbEFKbW1OTzFiQT09

At the core of recurrent neural circuit models is their recurrent connectivity matrix which encodes the interactions between the neurons. Many classical recurrent circuit models use connectivity matrices that are mathematically normal (eg, symmetric or ring networks have been used ubiquitously as the connectivity underlying attractor networks or models of visual cortex). While normal matrices are useful due to their mathematical tractability, they yield models that display a very limited range of neural dynamics. In light of experimental observations of rich neural dynamics in a variety of brain regions including visual cortex, frontal cortices, and motor cortex, a more general class of models that utilise non-normal connectivities have been developed over the last 15 years. In this journal club, we will give an overview of non-normal matrices and the neural dynamics they can exhibit and we will review 3 papers that make particular use of non-normal dynamics:

Memory without Feedback in a Neural Network, M. Goldman, Neuron 61:4, pp. 621-634.

Balanced Amplification: A New Mechanism of Selective Amplification of Neural Activity Patterns, B. Murphy and K. Miller, Neuron 61:4, pp. 635-648.

Network dynamics underlying OFF responses in the auditory cortex, G. Bondanelli, T. Deneux, B. Bathellier, and S. Ostojic, bioRxiv, 810655, 2020.