VASE: Variational Assorted Surprise Exploration for Reinforcement Learning

About this item

Title

VASE: Variational Assorted Surprise Exploration for Reinforcement Learning

Content partner

University of Otago

Collection

Otago University Research Archive

Description

Exploration in environments with continuous control and sparse rewards remains a key challenge in reinforcement learning (RL). One of the approaches to encourage more systematic and efficient exploration relies on surprise as an intrinsic reward for the agent. We introduce a new definition of surprise and its RL implementation named variational assorted surprise exploration (VASE). VASE uses a Bayesian neural network as a model of the environment dynamics and is trained using variational infe...

Read more

Format

Research paper

Research format

Scholarly text / Journal article

Thesis level

Article

Date created

2023-03

Creator

Xu, Haitao / Szymanski, Lech / McCane, Brendan

URL

https://hdl.handle.net/10523/23643

Related subjects

Bayes methods / Bayesian neural network (BNN) / Computational modeling / exploration / Neural networks / Predictive models / Probability distribution / Reinforcement learning / reinforcement learning (RL) / surprise / Training / variational inference

What is the copyright status of this item?

All Rights Reserved

This item is all rights reserved, which means you'll have to get permission from University of Otago before using it.

What can I do with this item?

You must always check with University of Otago to confirm the specific terms of use, but this is our understanding:

Non-infringing use

NZ Copyright law does not prevent every use of a copyright work. You should consider what you can and cannot do with a copyright work.

No sharing

You may not copy and/or share this item with others without further permission. This includes posting it on your blog, using it in a presentation, or any other public use.

No modifying

You are not allowed to adapt or remix this item into any other works.

No commercial use

You may not use this item commercially.