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

Traditional exploration methods in RL require agents to perform randomactions to find rewards. But these approaches struggle on sparse-reward domainslike Montezuma's Revenge where the probability that any random action sequenceleads to reward is extremely low. Recent algorithms have performed well on suchtasks by encouraging agents to visit new states or perform new actions inrelation to all prior training episodes (which we call across-trainingnovelty). But such algorithms do not consider whether an agent exhibitsintra-life novelty: doing something new within the current episode, regardlessof whether those behaviors have been performed in previous episodes. Wehypothesize that across-training novelty might discourage agents fromrevisiting initially non-rewarding states that could become important steppingstones later in training. We introduce Deep Curiosity Search (DeepCS), whichencourages intra-life exploration by rewarding agents for visiting as manydifferent states as possible within each episode, and show that DeepCS matchesthe performance of current state-of-the-art methods on Montezuma's Revenge. Wefurther show that DeepCS improves exploration on Amidar, Freeway, Gravitar, andTutankham (many of which are hard exploration games). Surprisingly, DeepCSdoubles A2C performance on Seaquest, a game we would not have expected tobenefit from intra-life exploration because the arena is small and alreadyeasily navigated by naive exploration techniques. In one run, DeepCS achieves amaximum training score of 80,000 points on Seaquest, higher than any methodsother than Ape-X. The strong performance of DeepCS on these sparse- anddense-reward tasks suggests that encouraging intra-life novelty is aninteresting, new approach for improving performance in Deep RL and motivatesfurther research into hybridizing across-training and intra-life explorationmethods.

Deep Curiosity Search: Intra-Life Exploration Can Improve Performance on Challenging Deep Reinforcement Learning Problems