Alladi Ramakrishnan Hall

A Deep Dive into Deep Learning Part 2: The new physics of learning

Anuran Pal

IMSc Chennai

This talk and its prequel (on Oct 25) will look in detail at the papers that together led to the ``physics revolution'' in AI, fetching John J. Hopfield and Geoffrey E. Hinton the Physics Nobel Prize this year. This second talk will focus on how Hinton extended the Hopfield model by introducing hidden nodes that significantly increased the capability of the earlier associative memory model. However, its training proved to be computationally extremely challenging, leading to the development of the Restricted Boltzmann Machine that allows connections only between visible and hidden nodes of the network. The improved performance of this bipartite structure eventually gave rise to the first deep neural network models - culminating in the Large Language Models of today.