English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

We analyze the hidden activations of neural network policies of deepreinforcement learning (RL) agents and show, empirically, that it's possible toknow a priori if a state representation will lend itself to fast learning. RLagents in high-dimensional states have two main learning burdens: (1) to learnan action-selection policy and (2) to learn to discern between useful andnon-useful information in a given state. By learning a latent representation ofthese high-dimensional states with an auxiliary model, the latter burden iseffectively removed, thereby leading to accelerated training progress. Weexamine this phenomenon across tasks in the PyBullet Kuka environment, where anagent must learn to control a robotic gripper to pick up an object. Ouranalysis reveals how neural network policies learn to organize their internalrepresentation of the state space throughout training. The results from thisanalysis provide three main insights into how deep RL agents learn. First, awell-organized internal representation within the policy network is aprerequisite to learning good action-selection. Second, a poor initialrepresentation can cause an unrecoverable collapse within a policy network.Third, a good initial representation allows an agent's policy network toorganize its internal representation even before any training begins.

Analyzing the Hidden Activations of Deep Policy Networks: Why Representation Matters