The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
Sabe-se que as Deep Q Networks comumente usadas superestimam os valores de ação sob certas condições. Também está comprovado que superestimações prejudicam o desempenho, podendo causar instabilidade e divergência de aprendizagem. Neste artigo, apresentamos o algoritmo Deep Sarsa and Q Networks (DSQN), que pode ser considerado um aprimoramento do algoritmo Deep Q Networks. Primeiro, o algoritmo DSQN aproveita a repetição da experiência e as técnicas de rede alvo nas Deep Q Networks para melhorar a estabilidade das redes neurais. Em segundo lugar, o estimador duplo é utilizado para Q-learning para reduzir superestimações. Especialmente, apresentamos o aprendizado Sarsa às Deep Q Networks para remover ainda mais as superestimações. Finalmente, o algoritmo DSQN é avaliado em tarefas de balanceamento de carrinho, carro de montanha e controle de lunarlander do OpenAI Gym. Os resultados da avaliação empírica mostram que o método proposto leva à redução de superestimações, a um processo de aprendizagem mais estável e a um melhor desempenho.
Zhi-xiong XU
PLA University of Science and Technology
Lei CAO
PLA University of Science and Technology
Xi-liang CHEN
PLA University of Science and Technology
Chen-xi LI
PLA University of Science and Technology
Yong-liang ZHANG
PLA University of Science and Technology
Jun LAI
PLA University of Science and Technology
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copiar
Zhi-xiong XU, Lei CAO, Xi-liang CHEN, Chen-xi LI, Yong-liang ZHANG, Jun LAI, "Deep Reinforcement Learning with Sarsa and Q-Learning: A Hybrid Approach" in IEICE TRANSACTIONS on Information,
vol. E101-D, no. 9, pp. 2315-2322, September 2018, doi: 10.1587/transinf.2017EDP7278.
Abstract: The commonly used Deep Q Networks is known to overestimate action values under certain conditions. It's also proved that overestimations do harm to performance, which might cause instability and divergence of learning. In this paper, we present the Deep Sarsa and Q Networks (DSQN) algorithm, which can considered as an enhancement to the Deep Q Networks algorithm. First, DSQN algorithm takes advantage of the experience replay and target network techniques in Deep Q Networks to improve the stability of neural networks. Second, double estimator is utilized for Q-learning to reduce overestimations. Especially, we introduce Sarsa learning to Deep Q Networks for removing overestimations further. Finally, DSQN algorithm is evaluated on cart-pole balancing, mountain car and lunarlander control task from the OpenAI Gym. The empirical evaluation results show that the proposed method leads to reduced overestimations, more stable learning process and improved performance.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2017EDP7278/_p
Copiar
@ARTICLE{e101-d_9_2315,
author={Zhi-xiong XU, Lei CAO, Xi-liang CHEN, Chen-xi LI, Yong-liang ZHANG, Jun LAI, },
journal={IEICE TRANSACTIONS on Information},
title={Deep Reinforcement Learning with Sarsa and Q-Learning: A Hybrid Approach},
year={2018},
volume={E101-D},
number={9},
pages={2315-2322},
abstract={The commonly used Deep Q Networks is known to overestimate action values under certain conditions. It's also proved that overestimations do harm to performance, which might cause instability and divergence of learning. In this paper, we present the Deep Sarsa and Q Networks (DSQN) algorithm, which can considered as an enhancement to the Deep Q Networks algorithm. First, DSQN algorithm takes advantage of the experience replay and target network techniques in Deep Q Networks to improve the stability of neural networks. Second, double estimator is utilized for Q-learning to reduce overestimations. Especially, we introduce Sarsa learning to Deep Q Networks for removing overestimations further. Finally, DSQN algorithm is evaluated on cart-pole balancing, mountain car and lunarlander control task from the OpenAI Gym. The empirical evaluation results show that the proposed method leads to reduced overestimations, more stable learning process and improved performance.},
keywords={},
doi={10.1587/transinf.2017EDP7278},
ISSN={1745-1361},
month={September},}
Copiar
TY - JOUR
TI - Deep Reinforcement Learning with Sarsa and Q-Learning: A Hybrid Approach
T2 - IEICE TRANSACTIONS on Information
SP - 2315
EP - 2322
AU - Zhi-xiong XU
AU - Lei CAO
AU - Xi-liang CHEN
AU - Chen-xi LI
AU - Yong-liang ZHANG
AU - Jun LAI
PY - 2018
DO - 10.1587/transinf.2017EDP7278
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E101-D
IS - 9
JA - IEICE TRANSACTIONS on Information
Y1 - September 2018
AB - The commonly used Deep Q Networks is known to overestimate action values under certain conditions. It's also proved that overestimations do harm to performance, which might cause instability and divergence of learning. In this paper, we present the Deep Sarsa and Q Networks (DSQN) algorithm, which can considered as an enhancement to the Deep Q Networks algorithm. First, DSQN algorithm takes advantage of the experience replay and target network techniques in Deep Q Networks to improve the stability of neural networks. Second, double estimator is utilized for Q-learning to reduce overestimations. Especially, we introduce Sarsa learning to Deep Q Networks for removing overestimations further. Finally, DSQN algorithm is evaluated on cart-pole balancing, mountain car and lunarlander control task from the OpenAI Gym. The empirical evaluation results show that the proposed method leads to reduced overestimations, more stable learning process and improved performance.
ER -