A Finite-Time Analysis of Q-Learning with Neural Network Function Approximation

Abstract

Q-learning with neural network function approximation (neural Q-learning for short) is among the most prevalent deep reinforcement learning algorithms. Despite its empirical success, the non-asymptotic convergence rate of neural Q-learning remains virtually unknown. In this paper, we present a finite-time analysis of a neural Q-learning algorithm, where the data are generated from a Markov decision process and the action-value function is approximated by a deep ReLU neural network. We prove that neural Q-learning finds the optimal policy with $O(1/\sqrt{T})$ convergence rate if the neural function approximator is sufficiently overparameterized, where $T$ is the number of iterations. To our best knowledge, our result is the first finite-time analysis of neural Q-learning under non-i.i.d. data assumption.

22 pages, 1 table. This version simplifies the proof and improves the presentation

References in corpus (11)

• Deep Learning in Neural Networks: An Overview
• Dueling Network Architectures for Deep Reinforcement Learning
• Neural Tangent Kernel: Convergence and Generalization in Neural Networks
• Safe, Multi-Agent, Reinforcement Learning for Autonomous Driving
• A Theoretical Analysis of Deep Q-Learning
• Finite-Sample Analysis of Proximal Gradient TD Algorithms
• Generalization Error Bounds of Gradient Descent for Learning Over-parameterized Deep ReLU Networks
• Finite-Time Error Bounds For Linear Stochastic Approximation and TD Learning
• Characterizing the Exact Behaviors of Temporal Difference Learning Algorithms Using Markov Jump Linear System Theory
• Gram-Gauss-Newton Method: Learning Overparameterized Neural Networks for Regression Problems
• Finite-Sample Analysis of Nonlinear Stochastic Approximation with Applications in Reinforcement Learning

Cited by in corpus (7)

• A Theoretical Analysis of Deep Q-Learning
• A Finite Time Analysis of Two Time-Scale Actor Critic Methods
• Finite-Time Analysis for Double Q-learning
• Single-Timescale Stochastic Nonconvex-Concave Optimization for Smooth Nonlinear TD Learning
• Momentum Q-learning with Finite-Sample Convergence Guarantee
• Implicit Under-Parameterization Inhibits Data-Efficient Deep Reinforcement Learning
• Provably Efficient Neural Estimation of Structural Equation Model: An Adversarial Approach