Applying Policy Iteration for Training Recurrent Neural Networks

TL;DR

This paper introduces a policy iteration-based algorithm for training recurrent neural networks by modeling the learning process as a nonlinear least-squares minimization, establishing a connection to reinforcement learning.

Contribution

It presents a novel policy iteration approach for RNN training, leveraging reinforcement learning principles to ensure convergence and improve learning efficiency.

Findings

The proposed method converges reliably under certain assumptions.

RNN training can be effectively framed within reinforcement learning.

The approach offers a new perspective on optimizing recurrent neural networks.

Abstract

Recurrent neural networks are often used for learning time-series data. Based on a few assumptions we model this learning task as a minimization problem of a nonlinear least-squares cost function. The special structure of the cost function allows us to build a connection to reinforcement learning. We exploit this connection and derive a convergent, policy iteration-based algorithm. Furthermore, we argue that RNN training can be fit naturally into the reinforcement learning framework.