We are hosting a series of Crick Institute-sponsored neuroscience talks at Imperial & Crick on how to get experiment and theory working together better. We think it’s an exciting time to tackle this question seriously thanks to the growth of "NeuroAI" and large scale experimental projects. Each session will have a talk around these topics and an extended networking / group discussion on the questions raised (with free food provided). We will write up a position paper based on this experience and would welcome contributions.
The first talk is next Thursday (19th March) from 4.30pm at Imperial College’s White City campus.
Speaker: Marcus Ghosh
Title: A taxonomy of recurrence
Abstract: How does the structure of a neural network shape its function? In this talk I will introduce partially recurrent neural networks (pRNNs): a model in which a set of connection pathways can be combined combinatorially to generate a complete taxonomy of architectures between feedforward and fully recurrent. I will then present two functional explorations across these structures, by viewing them as: linear dynamical systems and nonlinear task trained networks. These results will launch us into a discussion on the value of toy models in neuroscience.
The talk will be around 30m long with a focus on discussions afterwards.
For a bit of extra context, you might find it interesting to read Marcus’ recent short article in The Transmitter on the value of ‘toy models’ in neuroscience:
Registration for the event is free but required:
https://www.eventbrite.co.uk/e/understanding-the-brain-in-theory-and-practice-tickets-1983856488160
Future events in the series:
Hope to see you there!
Dan Goodman