Mathematics Colloquia and Seminars

Return to Colloquia & Seminar listing

Many networks in the brain are sparse, and the brain prunes away synapses during development and learning. What properties does it manage to preserve through this pruning process? How does it know which synapses are most redundant? I'll present a biologically-plausible, local, unsupervised learning rule which manages to nearly preserve the entire spectrum of a class of networks (and hence the dynamics governed by such networks) while reducing its number of edges from O(n^2) to O(nlogn). Relying solely on synaptic weights and noise-driven covariances of the network, our learning rule repurposes noise in our system, taking what is often considered an irritant and transforming it into a tool for computation. This work is done in collaboration with Rishidev Chaudhuri.

Seminars this quarter will be online on Zoom. Please see the math bio seminar series email list or contact the organizers for link and password.