Modular Multi-Objective Deep Reinforcement Learning with Decision Values
Journal Title: Annals of Computer Science and Information Systems - Year 2018, Vol 15, Issue
Abstract
In this work we present a method for using Deep Q-Networks (DQNs) in multi-objective environments. Deep Q-Networks provide remarkable performance in single objective problems learning from high-level visual state representations. However, in many scenarios (e.g in robotics, games), the agent needs to pursue multiple objectives simultaneously. We propose an architecture in which separate DQNs are used to control the agent's behaviour with respect to particular objectives. In this architecture we introduce decision values to improve the scalarization of multiple DQNs into a single action. Our architecture enables the decomposition of the agent's behaviour into controllable and replaceable sub-behaviours learned by distinct modules. Moreover, it allows to change the priorities of particular objectives post-learning while preserving the overall performance of the agent. To evaluate our solution we used a game-like simulator in which an agent - provided with high-level visual input - pursues multiple objectives in a 2D world.
Authors and Affiliations
Tomasz Tajmajer
Keywords
Related Articles
- EP ID EP569798
- DOI 10.15439/2018F231
- Views 20
- Downloads 0
Analysis of inter-channel dependencies in audio lossless block coding
In this paper the basics of data predictive modeling (using the method of minimization mean square error) for lossless audio compression are presented. The described research focuses on inter-channel analysis and setting...
Design of models for the tokenization of electric power industry basing on the blockchain technology
The problem of implementing modern technologies into the electric power industry is quite relevant in the world. The article considers the models of decentralized platforms providing services for energy distribution and...
Vehicular Ad-Hoc Network for Smart Cities
The rapid increase in urban population is alleviating various kinds of problems such as long hours traffic-jams, pollution which is making cities life insecure and non-livable. The notion of a smart city is proposed to i...
Parallelizing the code of the Fokker-Planck equation solution by stochastic approach in Julia programming language
Presenting a reliable physical simulation requires very often use of the supercomputers and models run for many days or weeks. The numerical computing is divided into two groups. One uses highly efficient low-level langu...
News articles similarity for automatic media bias detection in Polish news portals
Digital media have enormous impact on the public opinion. In the ideal world the news in public media should be presented in a fair and impartial way. In practice the information presented in digital media is often biase...