A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Reinforcement learning (RL) has demonstrated impressive results in various fields; however, its performance can be significantly hindered when the training and testing environments differ. Domain adaptation (DA) techniques aim to bridge this gap by moving knowledge between domains. This paper presents a thorough and systematic study of DA in RL. We review and categorize existing approaches for DA in RL, including model-free and model-based. We examine the drawbacks associated with each approach, such as sample inefficiency and generalization issues. Furthermore, we explore various strategies used in DA, such as feature adaptation, reward shaping, and data augmentation. We provide insights into the benefits and drawbacks of different techniques and propose future research directions for enhancing DA in RL. Through this study, the goal is to offer comprehensive insight into the current state of DA in RL and contribute to developing more robust and adaptable RL algorithms.
Reinforcement learning (RL) has demonstrated impressive results in various fields; however, its performance can be significantly hindered when the training and testing environments differ. Domain adaptation (DA) techniques aim to bridge this gap by moving knowledge between domains. This paper presents a thorough and systematic study of DA in RL. We review and categorize existing approaches for DA in RL, including model-free and model-based. We examine the drawbacks associated with each approach, such as sample inefficiency and generalization issues. Furthermore, we explore various strategies used in DA, such as feature adaptation, reward shaping, and data augmentation. We provide insights into the benefits and drawbacks of different techniques and propose future research directions for enhancing DA in RL. Through this study, the goal is to offer comprehensive insight into the current state of DA in RL and contribute to developing more robust and adaptable RL algorithms.
Domain Adaptation in Reinforcement Learning: Approaches, Limitations, and Future Directions
J. Inst. Eng. India Ser. B
Wang, Bin (author)
Journal of The Institution of Engineers (India): Series B ; 105 ; 1223-1240
2024-10-01
18 pages
Article (Journal)
Electronic Resource
English
Domain Adaptation in Reinforcement Learning: Approaches, Limitations, and Future Directions
| Springer Verlag | 2024
Soil reinforcement applications: Promising directions for the future
| British Library Conference Proceedings | 1996
Functionalized carbon nanotubes in bio-world: Applications, limitations and future directions
| British Library Online Contents | 2017
Adaptation Pathways in Coastal Case Studies: Lessons Learned and Future Directions
| Taylor & Francis Verlag | 2017