Stop Regressing: Training Value Functions via Classification for Scalable Deep RL

TL;DR

This paper proposes replacing the traditional mean squared error regression loss with a categorical cross-entropy classification loss for training value functions in deep reinforcement learning, leading to improved scalability and performance across various domains.

Contribution

The authors demonstrate that using classification loss for value functions enhances scalability and performance in deep RL, outperforming regression-based methods in multiple challenging tasks.

Findings

Improved performance on Atari 2600 games with SoftMoEs.

State-of-the-art results in robotic manipulation with Q-transformers.

Successful application to Chess and Wordle with high-capacity Transformers.

Abstract

Value functions are a central component of deep reinforcement learning (RL). These functions, parameterized by neural networks, are trained using a mean squared error regression objective to match bootstrapped target values. However, scaling value-based RL methods that use regression to large networks, such as high-capacity Transformers, has proven challenging. This difficulty is in stark contrast to supervised learning: by leveraging a cross-entropy classification loss, supervised methods have scaled reliably to massive networks. Observing this discrepancy, in this paper, we investigate whether the scalability of deep RL can also be improved simply by using classification in place of regression for training value functions. We demonstrate that value functions trained with categorical cross-entropy significantly improves performance and scalability in a variety of domains. These include: single-task RL on Atari 2600 games with SoftMoEs, multi-task RL on Atari with large-scale ResNets, robotic manipulation with Q-transformers, playing Chess without search, and a language-agent Wordle task with high-capacity Transformers, achieving state-of-the-art results on these domains. Through careful analysis, we show that the benefits of categorical cross-entropy primarily stem from its ability to mitigate issues inherent to value-based RL, such as noisy targets and non-stationarity. Overall, we argue that a simple shift to training value functions with categorical cross-entropy can yield substantial improvements in the scalability of deep RL at little-to-no cost.