Fast Reinforcement Learning Using Asymmetric Probability Density Function

Kosuke Umesako; Masanao Obayashi; Kunikazu Kobayashi

doi:10.11499/sicep.2002.0.171.0

SICE Annual Conference 2002

DOI https://doi.org/10.11499/sicep.2002.0.171.0

Conference information

Host: The Society of Instrument and Control Engineers

Co-host: IEEE/Industiral Electronic Society, IEEE/Robotics and Automation Society, IEEE/Control System Society

Fast Reinforcement Learning Using Asymmetric Probability Density Function

Kosuke Umesako, Masanao Obayashi, Kunikazu Kobayashi

Author information

Keywords: reinforcement learning, acceleration of learning, asymmetric probability density function, temporal difference method, stochastic gradient ascent method, neural network

CONFERENCE PROCEEDINGS FREE ACCESS

Pages 171

Details

Abstract

We propose an asymmetric probability density function (PDF) to select an effective action on reinforcement learning (RL). The proposed method utilizing the information of search direction enables RL to reduce the number of trials. Furthermore, the proposed method can be applied easily to various methods of RL, for example, actor-critic, stochastic gradient ascent method. The performance of our proposed method is demonstrated by computer simulations.

Corresponding author

Register with J-STAGE for free!