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

In recent years deep neural networks have been successfully applied to thedomains of reinforcement learning\cite{bengio2009learning,krizhevsky2012imagenet,hinton2006reducing}. Deepreinforcement learning \cite{mnih2015human} is reported to have the advantageof learning effective policies directly from high-dimensional sensory inputsover traditional agents. However, within the scope of the literature, there isno fundamental change or improvement on the existing training framework. Herewe propose a novel training framework that is conceptually comprehensible andpotentially easy to be generalized to all feasible algorithms for reinforcementlearning. We employ Monte-carlo sampling to achieve raw data inputs, and trainthem in batch to achieve Markov decision process sequences and synchronouslyupdate the network parameters instead of experience replay. This trainingframework proves to optimize the unbiased approximation of loss function whoseestimation exactly matches the real probability distribution data inputsfollow, and thus have overwhelming advantages of sample efficiency andconvergence rate over existing deep reinforcement learning after evaluating iton both discrete action spaces and continuous control problems. Besides, wepropose several algorithms embedded with our new framework to deal with typicaldiscrete and continuous scenarios. These algorithms prove to be far moreefficient than their original versions under the framework of deepreinforcement learning, and provide examples for existing and future algorithmsto generalize to our new framework.

Unbiased Deep Reinforcement Learning: A General Training Framework for Existing and Future Algorithms