ScholarWorks@숙명여자대학교

초록

Reinforcement learning (RL) has proven to be an effective solution for addressing sequential decision-making problems. However, its applicability is often limited in real-world scenarios where the interaction with the environment is restricted or prohibited. To overcome this challenge, offline RL has been proposed, which seeks to optimize actions based on the datasets gathered from the environment. Recent advancements in of generative artificial intelligence (GAI) have been integrated with offline RL, resulting in GAI-based offline RL. This approach leverages the capabilities of GAI to approximate arbitrary probability distribution, allowing it to mimic the distribution of the expert datasets. This paper conducts a comprehensive study of the theoretical background and properties of the diffusion-based RL algorithms, and analyzes their performance in various RL environments. Furthermore, it outlines the future directions for research on diffusion-based offline RL based on this analysis.

키워드