Analysis and Optimisation of Bellman Residual Errors with Neural Function Approximation
This work addresses the problem of unreliable and inefficient optimization in deep reinforcement learning for researchers and practitioners, offering an incremental improvement over existing methods.
The paper tackles the challenge of minimizing the Mean Squared Bellman Error (MSBE) in deep reinforcement learning by analyzing it from a smooth optimization perspective and developing an Approximate Newton's algorithm. The result includes a Gauss Newton Residual Gradient algorithm that achieves local quadratic convergence to a global minimum in discrete settings and demonstrates feasibility in continuous control problems.
Recent development of Deep Reinforcement Learning (DRL) has demonstrated superior performance of neural networks in solving challenging problems with large or even continuous state spaces. One specific approach is to deploy neural networks to approximate value functions by minimising the Mean Squared Bellman Error (MSBE) function. Despite great successes of DRL, development of reliable and efficient numerical algorithms to minimise the MSBE is still of great scientific interest and practical demand. Such a challenge is partially due to the underlying optimisation problem being highly non-convex or using incomplete gradient information as done in Semi-Gradient algorithms. In this work, we analyse the MSBE from a smooth optimisation perspective and develop an efficient Approximate Newton's algorithm. First, we conduct a critical point analysis of the error function and provide technical insights on optimisation and design choices for neural networks. When the existence of global minima is assumed and the objective fulfils certain conditions, suboptimal local minima can be avoided when using over-parametrised neural networks. We construct a Gauss Newton Residual Gradient algorithm based on the analysis in two variations. The first variation applies to discrete state spaces and exact learning. We confirm theoretical properties of this algorithm such as being locally quadratically convergent to a global minimum numerically. The second employs sampling and can be used in the continuous setting. We demonstrate feasibility and generalisation capabilities of the proposed algorithm empirically using continuous control problems and provide a numerical verification of our critical point analysis. We outline the difficulties of combining Semi-Gradient approaches with Hessian information. To benefit from second-order information complete derivatives of the MSBE must be considered during training.