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/

Zendy Plus

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

Zendy Tools

Zendy Open

Studies have shown evolution strategies (ES) to be a promising approach forreinforcement learning (RL) with deep neural networks. However, the issue ofhigh sample complexity persists in applications of ES to deep RL over longhorizons. This paper is the first to address the shortcoming of today's methodsvia a novel neuroevolutionary multitasking (NuEMT) algorithm, designed totransfer information from a set of auxiliary tasks (of short episode length) tothe target (full length) RL task at hand. The auxiliary tasks, extracted fromthe target, allow an agent to update and quickly evaluate policies on shortertime horizons. The evolved skills are then transferred to guide the longer andharder task towards an optimal policy. We demonstrate that the NuEMT algorithmachieves data-efficient evolutionary RL, reducing expensive agent-environmentinteraction data requirements. Our key algorithmic contribution in this settingis to introduce, for the first time, a multitask skills transfer mechanismbased on the statistical importance sampling technique. In addition, anadaptive resource allocation strategy is utilized to assign computationalresources to auxiliary tasks based on their gleaned usefulness. Experiments ona range of continuous control tasks from the OpenAI Gym confirm that ourproposed algorithm is efficient compared to recent ES baselines.

Multitask Neuroevolution for Reinforcement Learning with Long and Short Episodes