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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Risk-sensitive reinforcement learning (RL) has garnered significant attentionin recent years due to the growing interest in deploying RL agents inreal-world scenarios. A critical aspect of risk awareness involves modelinghighly rare risk events (rewards) that could potentially lead to catastrophicoutcomes. These infrequent occurrences present a formidable challenge fordata-driven methods aiming to capture such risky events accurately. Whilerisk-aware RL techniques do exist, their level of risk aversion heavily relieson the precision of the state-action value function estimation when modelingthese rare occurrences. Our work proposes to enhance the resilience of RLagents when faced with very rare and risky events by focusing on refining thepredictions of the extreme values predicted by the state-action value functiondistribution. To achieve this, we formulate the extreme values of thestate-action value function distribution as parameterized distributions,drawing inspiration from the principles of extreme value theory (EVT). Thisapproach effectively addresses the issue of infrequent occurrence by leveragingEVT-based parameterization. Importantly, we theoretically demonstrate theadvantages of employing these parameterized distributions in contrast to otherrisk-averse algorithms. Our evaluations show that the proposed methodoutperforms other risk averse RL algorithms on a diverse range of benchmarktasks, each encompassing distinct risk scenarios.

Extreme Risk Mitigation in Reinforcement Learning using Extreme Value Theory