Q-learning with heterogeneous update strategy

Article Computer Science, Artificial Intelligence

Sub-AVG: Overestimation reduction for cooperative multi-agent reinforcement learning

Haolin Wu et al.

Summary: Decomposing the centralized joint action value into per-agent individual action value is attractive in cooperative multi-agent reinforcement learning. However, the Q-learning-based method suffers from overestimation. This paper presents a solution called Sub-AVG, which eliminates excessive overestimation errors by using a lower update target.

NEUROCOMPUTING (2022)

Add to Collection

Article Mathematics, Interdisciplinary Applications

Markov Chain Monte Carlo in Practice

Galin L. Jones et al.

Summary: This article reviews methods for assessing the reliability of Markov chain Monte Carlo (MCMC) simulation results, with a focus on those most useful in practical settings.

ANNUAL REVIEW OF STATISTICS AND ITS APPLICATION (2022)

Add to Collection

Article Energy & Fuels

Energy saving evaluation of an energy efficient data center using a model-free reinforcement learning approach

Muhammad Haiqal Bin Mahbod et al.

Summary: In this study, a hybrid data center model is built using a deep reinforcement learning algorithm to reduce energy consumption in tropical climates. The results show that using a floating setpoint and lowering the temperature value can achieve significant energy savings, with the reduction of server fan usage being the main contributor.

APPLIED ENERGY (2022)

Add to Collection

Article Computer Science, Information Systems

A Swapping Target Q-Value Technique for Data Augmentation in Offline Reinforcement Learning

Ho-Taek Joo et al.

Summary: This study introduces a novel data augmentation technique called Swapping Target Q-Value (SQV) to enhance offline RL algorithms and improve pixel-based learning. By matching the Q-values of transformed images with the target Q-values of original images, and considering similar states as the same and different states as more distinct, the performance of the method is observed to significantly increase in the Atari 2600 game domain.

IEEE ACCESS (2022)

Add to Collection

Article Engineering, Civil