NDNS+ WORKSHOP June 22-23, 2020 ONLINE MEETING
Date:
22-23 June 2020
Speaker: Prof. Peter Ashwin - Exeter University
Title: Computational properties of network attractors
Abstract:
Both natural and artificial neural networks perform computational tasks via their dynamic response to time-varying inputs. This talk will discuss recent work on excitable network attractors, where stable patterns of the network are nonetheless susceptible to small perturbations in certain directions. Such network attractors, which can be created at bifurcation from heteroclinic attractors, can provide a framework that explains how responses may be encoded and used and perform finite state computation, at least for artificial recurrent neural networks. This research is in collaboration with Claire Postlethwaite, Andrea Ceni and Lorenzo Livi.