Continual Multi-Objective Reinforcement Learning via Reward Model Rehearsal
Abstract
Languages for Knowledge Representation and Reasoning, such as ASP, CP, and SMT, excel at solving some complex problems, but encoding them into a higher-level language may be more profitable, leaving these formalisms as targets for solving. Recent studies aim to convert controlled natural languages into formal representations, yet these solutions are often tailored to specific languages and require significant effort. This paper introduces a general framework that generates grammars for target representation languages, enabling the translation of problems stated in CNL into formal representations. The related system, CNLWizard, offers a flexible, high-level approach to defining desired grammars, significantly reducing the time and effort needed to create custom grammars. Finally, we demonstrate the system's effectiveness through an experimental analysis.
Cite
Text
Li et al. "Continual Multi-Objective Reinforcement Learning via Reward Model Rehearsal." International Joint Conference on Artificial Intelligence, 2024. doi:10.24963/ijcai.2024/490Markdown
[Li et al. "Continual Multi-Objective Reinforcement Learning via Reward Model Rehearsal." International Joint Conference on Artificial Intelligence, 2024.](https://mlanthology.org/ijcai/2024/li2024ijcai-continual/) doi:10.24963/ijcai.2024/490BibTeX
@inproceedings{li2024ijcai-continual,
title = {{Continual Multi-Objective Reinforcement Learning via Reward Model Rehearsal}},
author = {Li, Lihe and Chen, Ruotong and Zhang, Ziqian and Wu, Zhichao and Li, Yi-Chen and Guan, Cong and Yu, Yang and Yuan, Lei},
booktitle = {International Joint Conference on Artificial Intelligence},
year = {2024},
pages = {4434-4442},
doi = {10.24963/ijcai.2024/490},
url = {https://mlanthology.org/ijcai/2024/li2024ijcai-continual/}
}