SimBa: Simplicity Bias for Scaling Up Parameters in Deep Reinforcement Learning

TL;DR

SimBa introduces a simple architecture that scales up parameters in deep reinforcement learning by embedding a simplicity bias, leading to improved sample efficiency and state-of-the-art performance across various environments.

Contribution

The paper proposes SimBa, a novel architecture that effectively scales parameters in deep RL through simplicity bias, enhancing performance and efficiency.

Findings

Improves sample efficiency across multiple RL algorithms.

Matches or surpasses state-of-the-art performance.

Demonstrates broad applicability across environments.

Abstract

Recent advances in CV and NLP have been largely driven by scaling up the number of network parameters, despite traditional theories suggesting that larger networks are prone to overfitting. These large networks avoid overfitting by integrating components that induce a simplicity bias, guiding models toward simple and generalizable solutions. However, in deep RL, designing and scaling up networks have been less explored. Motivated by this opportunity, we present SimBa, an architecture designed to scale up parameters in deep RL by injecting a simplicity bias. SimBa consists of three components: (i) an observation normalization layer that standardizes inputs with running statistics, (ii) a residual feedforward block to provide a linear pathway from the input to output, and (iii) a layer normalization to control feature magnitudes. By scaling up parameters with SimBa, the sample efficiency of various deep RL algorithms-including off-policy, on-policy, and unsupervised methods-is consistently improved. Moreover, solely by integrating SimBa architecture into SAC, it matches or surpasses state-of-the-art deep RL methods with high computational efficiency across DMC, MyoSuite, and HumanoidBench. These results demonstrate SimBa's broad applicability and effectiveness across diverse RL algorithms and environments.