Name: Loss of Plasticity in Deep Continual Learning
Start: 2024-08-22T18:00:00-07:00
End: 2024-08-22T20:00:00-07:00

**[FounderCulture](http://founderculture.net/), [Storytell.ai](http://Storytell.ai) and [UpHonest Capital](https://go.founderculture.net/uphonest) have teamed up to create a Meetup series called "Building The Future of AI."**

**🟢 This event will be live in San Mateo, CA -- [also be live-streamed on this Zoom link](https://go.storytell.ai/drodio-zoom)** if you can't join in person. The Zoom will start at 6:30pm PT.

**Join us for an AI research paper technical deep dive:**
Rich Sutton’s student [Shibhansh Dohare](https://ca.linkedin.com/in/shibhansh-dohare) will present his paper, “Loss of Plasticity in Deep Continual Learning”, followed by a group discussion moderated by [DROdio](http://linkedin.com/in/drodio).

**Here is an abstract of the paper and presentation:**
===

Artificial neural networks, deep-learning methods and the backpropagation algorithm1 form the foundation of modern machine learning and artificial intelligence. These methods are almost always used in two phases, one in which the weights of the network are updated and one in which the weights are held constant while the network is used or evaluated. This contrasts with natural learning and many applications, which require continual learning.

It has been unclear whether deep-learning methods work in continual-learning settings. Here we show that they do not—that standard deep-learning methods gradually lose plasticity in continual-learning settings until they learn no better than a shallow network. We show such loss of plasticity using the classic ImageNet dataset and reinforcement-learning problems across a wide range of variations in the network and the learning algorithm.

Plasticity is maintained indefinitely only by algorithms that continually inject diversity into the network, such as our continual backpropagation algorithm, a variation of backpropagation in which a small fraction of less-used units are continually and randomly reinitialized.

Our results indicate that methods based on gradient descent are not enough—that sustained deep learning requires a random, non-gradient component to maintain variability and plasticity.

**🎁 FounderCulture is powered by our Patrons** -- [consider becoming a Patron to support our vision](http://www.patreon.com/FounderCulture) of improving the odds for Founders globally *(you'll get early access to RSVP to our events!)*

FounderCulture

Building the Future of AI

Technology

New Technology

Big Data

Founders

Artificial Intelligence Applications

AI and Society

Technology Startups

Artificial Intelligence Machine Learning Robotics

Artificial Intelligence

AI Algorithms

Shibhansh is a Ph.D. student in the RLAI lab at the University of Alberta, advised by Dr. Richard Sutton and Dr. Rupam Mahmood. They completed their undergraduate studies at IIT Kanpur, majoring in Computer Science and Engineering.

His long-term research goal is to understand the workings of the human mind, specifically to uncover the fundamental computational principles that give rise to it. In pursuit of this goal, he is working on various aspects of continual learning, deep learning, and reinforcement learning.

Shibhansh Dohare

Loss of Plasticity in Deep Continual Learning

WeWork Crossroads

Share

What does your past 12 months of Meetup reveal about you? It's time to relive your personal highlights!

Don't miss your <highlight>2025 recap!</highlight>

Building the Future of AI

Loss of Plasticity in Deep Continual Learning

Building the Future of AI

Details

Members are also interested in