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

The optimal policy of a reinforcement learning problem is often discontinuousand non-smooth. I.e., for two states with similar representations, theiroptimal policies can be significantly different. In this case, representing theentire policy with a function approximator (FA) with shared parameters for allstates maybe not desirable, as the generalization ability of parameters sharingmakes representing discontinuous, non-smooth policies difficult. A common wayto solve this problem, known as Mixture-of-Experts, is to represent the policyas the weighted sum of multiple components, where different components performwell on different parts of the state space. Following this idea and inspired bya recent work called advantage-weighted information maximization, we propose tolearn for each state weights of these components, so that they entail theinformation of the state itself and also the preferred action learned so farfor the state. The action preference is characterized via the advantagefunction. In this case, the weight of each component would only be large forcertain groups of states whose representations are similar and preferred actionrepresentations are also similar. Therefore each component is easy to berepresented. We call a policy parameterized in this way an Advantage WeightedMixture Policy (AWMP) and apply this idea to improve soft-actor-critic (SAC),one of the most competitive continuous control algorithm. Experimental resultsdemonstrate that SAC with AWMP clearly outperforms SAC in four commonly usedcontinuous control tasks and achieve stable performance across different randomseeds.

Off-policy Maximum Entropy Reinforcement Learning : Soft Actor-Critic with Advantage Weighted Mixture Policy(SAC-AWMP)