Improving the Performance of Deep Q Network in Decision Making Environment: Applying Multi-Head Attention into DQN

Ziyi Lu

2024

Abstract

Deep q-network (DQN) is a deep reinforcement learning algorithm. It has been used to solve value function-based reinforcement learning problems, and with deep neural networks, it can solve complex decision-making tasks. Multi-Head Attention is a mechanism that can capture various aspects of the input data in parallel, improving the model's capacity to recognize complex patterns in the data. Mahjong, being a traditional multi-player imperfect information card game, fits well in training and testing a decision-making model. This study introduces an application of combination of DQN and Multi-Head Attention mechanism, seeking to enhance the performance of DQN models by leveraging the benefits of multi-head attention. Through RL Card platform, which provides a variant of mahjong model that can be used for training, the paper applies the Multi-Head Attention with its original DQN agent for mahjong. The trained model shows a higher winning rate in the battle against the baseline model, demonstrating the promotion the combination brings to the mahjong agent model.

Paper Citation

in Harvard Style

Lu Z. (2024). Improving the Performance of Deep Q Network in Decision Making Environment: Applying Multi-Head Attention into DQN. In Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI; ISBN 978-989-758-713-9, SciTePress, pages 522-526. DOI: 10.5220/0012958200004508

in Bibtex Style

@conference{emiti24,
author={Ziyi Lu},
title={Improving the Performance of Deep Q Network in Decision Making Environment: Applying Multi-Head Attention into DQN},
booktitle={Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI},
year={2024},
pages={522-526},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012958200004508},
isbn={978-989-758-713-9},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI
TI - Improving the Performance of Deep Q Network in Decision Making Environment: Applying Multi-Head Attention into DQN
SN - 978-989-758-713-9
AU - Lu Z.
PY - 2024
SP - 522
EP - 526
DO - 10.5220/0012958200004508
PB - SciTePress