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 Vision Transformer architecture has shown to be competitive in thecomputer vision (CV) space where it has dethroned convolution-based networks inseveral benchmarks. Nevertheless, convolutional neural networks (CNN) remainthe preferential architecture for the representation module in reinforcementlearning. In this work, we study pretraining a Vision Transformer using severalstate-of-the-art self-supervised methods and assess the quality of the learnedrepresentations. To show the importance of the temporal dimension in thiscontext we propose an extension of VICReg to better capture temporal relationsbetween observations by adding a temporal order verification task. Our resultsshow that all methods are effective in learning useful representations andavoiding representational collapse for observations from Atari LearningEnvironment (ALE) which leads to improvements in data efficiency when weevaluated in reinforcement learning (RL). Moreover, the encoder pretrained withthe temporal order verification task shows the best results across allexperiments, with richer representations, more focused attention maps andsparser representation vectors throughout the layers of the encoder, whichshows the importance of exploring such similarity dimension. With this work, wehope to provide some insights into the representations learned by ViT during aself-supervised pretraining with observations from RL environments and whichproperties arise in the representations that lead to the best-performingagents. The source code will be available at:https://github.com/mgoulao/TOV-VICReg

Pretraining the Vision Transformer using self-supervised methods for vision based Deep Reinforcement Learning