+ "abstract": "With the rapid development of Large Language Models (LLMs), LLM-based agents have been widely adopted\nin various fields, becoming essential for autonomous decision-making and interactive tasks. However, current\nwork typically relies on prompt design or fine-tuning strategies applied to vanilla LLMs, which often leads\nto limited effectiveness or suboptimal performance in complex agent-related environments. Although LLM\noptimization techniques can improve model performance across many general tasks, they lack specialized\noptimization towards critical agent functionalities such as long-term planning, dynamic environmental\ninteraction, and complex decision-making. Although numerous recent studies have explored various strategies\nto optimize LLM-based agents for complex agent tasks, a systematic review summarizing and comparing these\nmethods from a holistic perspective is still lacking. In this survey, we provide a comprehensive review of LLM-\nbased agent optimization approaches, categorizing them into parameter-driven and parameter-free methods.\nWe first focus on parameter-driven optimization, covering fine-tuning-based optimization, reinforcement\nlearning-based optimization, and hybrid strategies, analyzing key aspects such as trajectory data construction,\nfine-tuning techniques, reward function design, and optimization algorithms. Additionally, we briefly discuss\nparameter-free strategies that optimize agent behavior through prompt engineering and external knowledge\nretrieval. Finally, we summarize the datasets and benchmarks used for evaluation and tuning, review key\napplications of LLM-based agents, and discuss major challenges and promising future directions. Our repository\nfor related references is available at https://github.com/YoungDubbyDu/LLM-Agent-Optimization."
0 commit comments