Author(s)

Runxi Kang

Corresponding Author

Runxi Kang

Abstract

Reinforcement Learning (RL) has achieved remarkable development in the history of artificial intelligence in the past decade. Especially from 2020 to 2025, its applications to CV and NLP have developed extremely quickly. Integrating deeply with deep learning techniques, RL has been applied to a series of challenging perception and decisions-making tasks. In CV, for example, RL is used to direct models to attend to important visual areas and to execute sequential perceptual tasks [1]. In NLP, it is employed to enhance text generation quality, train policies for dialog systems, and fine-tune large language models with reinforcement learning from human feedback (RLHF) to better match model output with human preference [2]. This survey comprehensively summarizes the major application areas and techniques of RL both in CV and NLP, shows some representative advances achieved in the past a few years, interprets the position and benefits of mainstream RL techniques applied to various tasks, explores existing challenges and shortages, and predicts future research trends.

Keywords

Reinforcement Learning, Computer Vision, Natural Language Processing