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

Training deep reinforcement learning (DRL) models usually requires highcomputation costs. Therefore, compressing DRL models possesses immensepotential for training acceleration and model deployment. However, existingmethods that generate small models mainly adopt the knowledgedistillation-based approach by iteratively training a dense network. As aresult, the training process still demands massive computing resources. Indeed,sparse training from scratch in DRL has not been well explored and isparticularly challenging due to non-stationarity in bootstrap training. In thiswork, we propose a novel sparse DRL training framework, "the RiggedReinforcement Learning Lottery" (RLx2), which builds upon gradient-basedtopology evolution and is capable of training a sparse DRL model based entirelyon a sparse network. Specifically, RLx2 introduces a novel multi-step TD targetmechanism with a dynamic-capacity replay buffer to achieve robust valuelearning and efficient topology exploration in sparse models. It also reachesstate-of-the-art sparse training performance in several tasks, showing7.5\times-20\times model compression with less than 3% performance degradationand up to 20\times and 50\times FLOPs reduction for training and inference,respectively.

RLx2: Training a Sparse Deep Reinforcement Learning Model from Scratch