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

This paper addresses the problem of maintaining safety during training inReinforcement Learning (RL), such that the safety constraint violations arebounded at any point during learning. In a variety of RL applications thesafety of the agent is particularly important, e.g. autonomous platforms orrobots that work in proximity of humans. As enforcing safety during trainingmight severely limit the agent's exploration, we propose here a newarchitecture that handles the trade-off between efficient progress and safetyduring exploration. As the exploration progresses, we update via Bayesianinference Dirichlet-Categorical models of the transition probabilities of theMarkov decision process that describes the environment dynamics. This paperproposes a way to approximate moments of belief about the risk associated tothe action selection policy. We construct those approximations, and prove theconvergence results. We propose a novel method for leveraging the expectationapproximations to derive an approximate bound on the confidence that the riskis below a certain level. This approach can be easily interleaved with RL andwe present experimental results to showcase the performance of the overallarchitecture.

Safeguarded Progress in Reinforcement Learning: Safe Bayesian Exploration for Control Policy Synthesis