Nikita Balagansky: Pruning with Catalyst

Data fest Online 2020 Catalyst Workshop track In the past few years, state-of-the-art architectures become more and more complex the number of parameters grows exponentially. But what if all networks are over-parametrized and more than a half parameters don’t influence the result. The process of removing connections between neurons called pruning. Moreover, some examples came from the biology field. For instance, the human brain is also over-parametrized in the first stages of growth, and we are learning through pruning unnecessary connections. In my presentation, I will tell you how to perform pruning of your network with the Catalyst framework and will briefly demonstrate what is “Lottery Ticket Hypothesis“ mean. Register and get access to the tracks: Join the community: