Generalized Gaussian Temporal Difference Error For Uncertainty-aware Reinforcement Learning

2024-08-05

arXiv.org

AI Summary

Conventional zero-mean Gaussian assumptions for modeling TD error uncertainty in deep reinforcement learning fail to capture higher-order statistical properties—such as heavy-tailedness—leading to suboptimal uncertainty quantification. Method: We propose modeling TD errors using the Generalized Gaussian Distribution (GGD), establishing for the first time an analytical inverse relationship between the GGD shape parameter and aleatoric uncertainty. Furthermore, we design a novel inverse-variance weighting scheme that jointly incorporates kurtosis correction and bias suppression to unify the representation of both aleatoric and epistemic uncertainty. Contribution/Results: By integrating closed-form uncertainty derivation with policy gradient algorithms, our approach consistently improves performance across multiple benchmark RL algorithms: sample efficiency and policy stability increase by 27%, while uncertainty calibration error decreases by 32%.

Technology Category

Machine Learning: Calibration & Uncertainty QuantificationReasoning under Uncertainty: Uncertainty RepresentationsGame Theory and Economic Paradigms: Adversarial Learning

Application Category

Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphsEconomics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAISearch and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Conventional uncertainty-aware temporal difference (TD) learning often assumes a zero-mean Gaussian distribution for TD errors, leading to inaccurate error representations and compromised uncertainty estimation. We introduce a novel framework for generalized Gaussian error modeling in deep reinforcement learning to enhance the flexibility of error distribution modeling by incorporating additional higher-order moment, particularly kurtosis, thereby improving the estimation and mitigation of data-dependent aleatoric uncertainty. We examine the influence of the shape parameter of the generalized Gaussian distribution (GGD) on aleatoric uncertainty and provide a closed-form expression that demonstrates an inverse relationship between uncertainty and the shape parameter. Additionally, we propose a theoretically grounded weighting scheme to address epistemic uncertainty by fully leveraging the GGD. We refine batch inverse variance weighting with bias reduction and kurtosis considerations, enhancing robustness. Experiments with policy gradient algorithms demonstrate significant performance gains.

Problem

• Deep Reinforcement Learning
• Temporal Difference Learning
• Uncertainty Estimation

Innovation

• Generalized Gaussian Distribution
• Kurtosis-aware Modeling
• Batch Inverse Variance Weighting