The Dormant Neuron Phenomenon in Deep Reinforcement Learning

We identify the dormant neuron phenomenon in deep reinforcement learning, where an agent’s network suffers from an increasing number of inactive neurons, thereby affecting network expressivity.

Ghada Sokar, Rishabh Agarwal, Pablo Samuel Castro*, Utku Evci*


This blogpost is a summary of our ICML 2023 paper. The code is available here. Many more results and analyses are available in the paper, so I encouraged you to check it out if interested!

The following figure gives a nice summary of the overall findings of our work (we are reporting the Interquantile Mean (IQM) as introduced in our Statistical Precipice NeurIPS'21 paper):

Metrics and continuity in reinforcement learning

In this work we investigate the notion of “state similarity” in Markov decision processes. This concept is central to generalization in RL with function approximation.

Our paper was published at AAAI'21.

Charline Le Lan, Marc G. Bellemare, and Pablo Samuel Castro

The text below was adapted from Charline’s twitter thread

In RL, we often deal with systems with large state spaces. We can’t exactly represent the value of each of these states and need some type of generalization. One way to do that is to look at structured representations in which similar states are assigned similar predictions.

Contrastive Behavioral Similarity Embeddings for Generalization in Reinforcement Learning

This paper was accepted as a spotlight at ICLR'21.

We propose a new metric and contrastive loss that comes equipped with theoretical and empirical results.

Jumping task

Policy Similarity Metric

We introduce the policy similarity metric (PSM) which is based on bisimulation metrics. In contrast to bisimulation metrics (which is built on reward differences), PSMs are built on differences in optimal policies.

Policy similarity metric

If we were to use this metric for policy transfer (as Doina Precup & I explored previously), we can upper-bound the difference between the optimal and the transferred policy:

2020 RL highlights

As part of TWiML ’s AI Rewind series, I was asked to provide a list of reinforcement learning papers that were highlights for me in 2020. It’s been a difficult year for pretty much everyone, but it’s heartening to see that despite all the difficulties, interesting research still came out.

Given the size and breadth of the reinforcement learning research, as well as the fact that I was asked to do this at the end of NeurIPS and right before my vacation, I decided to apply the following rules in the selection:

Autonomous navigation of stratospheric balloons using reinforcement learning

In this work we, quite literally, take reinforcement learning to new heights! Specifically, we use deep reinforcement learning to help control the navigation of stratospheric balloons, whose purpose is to deliver internet to areas with low connectivity. This project is an ongoing collaboration with Loon.

It’s been incredibly rewarding to see reinforcement learning deployed successfully in a real setting. It’s also been terrific to work alongside such fantastic co-authors:
Marc G. Bellemare, Salvatore Candido, Pablo Samuel Castro, Jun Gong, Marlos C. Machado, Subhodeep Moitra, Sameera S. Ponda, Ziyu Wang

GANterpretations

GANterpretations is an idea I published in this paper, which was accepted to the 4th Workshop on Machine Learning for Creativity and Design at NeurIPS 2020. The code is available here.

At a high level what it does is use the spectrogram of a piece of audio (from a video, for example) to “draw” a path in the latent space of a BigGAN.

The following video walks through the process:

GANs

GANs are generative models trained to reproduce images from a given dataset. The way GANs work is they are trained to learn a latent space $ Z\in\mathbb{R}^d $, where each point $ z\in Z $ generates a unique image $ G(z) $, where $ G $ is the generator of the GAN. When trained properly, these latent spaces are learned in a structured manner, where nearby points generate similar images.