Spectral Normalisation for Deep Reinforcement Learning: an Optimisation Perspective

by   Florin Gogianu, et al.

Most of the recent deep reinforcement learning advances take an RL-centric perspective and focus on refinements of the training objective. We diverge from this view and show we can recover the performance of these developments not by changing the objective, but by regularising the value-function estimator. Constraining the Lipschitz constant of a single layer using spectral normalisation is sufficient to elevate the performance of a Categorical-DQN agent to that of a more elaborated agent on the challenging Atari domain. We conduct ablation studies to disentangle the various effects normalisation has on the learning dynamics and show that is sufficient to modulate the parameter updates to recover most of the performance of spectral normalisation. These findings hint towards the need to also focus on the neural component and its learning dynamics to tackle the peculiarities of Deep Reinforcement Learning.


page 5

page 14

page 17

page 18

page 19

page 22

page 23

page 24


Neural Episodic Control

Deep reinforcement learning methods attain super-human performance in a ...

Abstraction for Deep Reinforcement Learning

We characterise the problem of abstraction in the context of deep reinfo...

Rainbow: Combining Improvements in Deep Reinforcement Learning

The deep reinforcement learning community has made several independent i...

Deep Q-Network with Proximal Iteration

We employ Proximal Iteration for value-function optimization in reinforc...

Deep Reinforcement Learning and the Deadly Triad

We know from reinforcement learning theory that temporal difference lear...

Equivalence Between Wasserstein and Value-Aware Model-based Reinforcement Learning

Learning a generative model is a key component of model-based reinforcem...

Attend2Pack: Bin Packing through Deep Reinforcement Learning with Attention

This paper seeks to tackle the bin packing problem (BPP) through a learn...