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

Count-based exploration algorithms are known to perform near-optimally whenused in conjunction with tabular reinforcement learning (RL) methods forsolving small discrete Markov decision processes (MDPs). It is generallythought that count-based methods cannot be applied in high-dimensional statespaces, since most states will only occur once. Recent deep RL explorationstrategies are able to deal with high-dimensional continuous state spacesthrough complex heuristics, often relying on optimism in the face ofuncertainty or intrinsic motivation. In this work, we describe a surprisingfinding: a simple generalization of the classic count-based approach can reachnear state-of-the-art performance on various high-dimensional and/or continuousdeep RL benchmarks. States are mapped to hash codes, which allows to counttheir occurrences with a hash table. These counts are then used to compute areward bonus according to the classic count-based exploration theory. We findthat simple hash functions can achieve surprisingly good results on manychallenging tasks. Furthermore, we show that a domain-dependent learned hashcode may further improve these results. Detailed analysis reveals importantaspects of a good hash function: 1) having appropriate granularity and 2)encoding information relevant to solving the MDP. This exploration strategyachieves near state-of-the-art performance on both continuous control tasks andAtari 2600 games, hence providing a simple yet powerful baseline for solvingMDPs that require considerable exploration.

#Exploration: A Study of Count-Based Exploration for Deep Reinforcement Learning