EvoDS: Self-Evolving Autonomous Data Science Agent with Skill Learning and Context Management
Researchers have developed EvoDS, a novel autonomous data science agent designed to overcome the limitations of static action sets and poor context management in current LLM agents. EvoDS employs an Autonomous Skill Acquisition mechanism to learn and reuse skills, alongside an Adaptive Context Compression strategy for efficient long-term context handling. This self-evolving agent has demonstrated a significant performance improvement, outperforming existing agents by an average of 28.9% across multiple benchmarks and eliminating out-of-token errors. AI
IMPACT Enhances autonomous data science capabilities by enabling agents to learn new skills and manage context more effectively.