Negative Learning Rates and P-Learning

TL;DR

This paper introduces a novel training method using negative learning rates for differentiable function approximators, applicable to regression tasks and direct policy learning in reinforcement learning.

Contribution

It proposes a new approach leveraging negative learning rates for training and demonstrates its use in reinforcement learning policy optimization.

Findings

Effective training with negative examples using negative learning rates

Application of the method to direct policy learning in reinforcement learning

Potential improvements in training efficiency or policy performance

Abstract

We present a method of training a differentiable function approximator for a regression task using negative examples. We effect this training using negative learning rates. We also show how this method can be used to perform direct policy learning in a reinforcement learning setting.