Heterogeneous Multi-Agent Reinforcement Learning with Attention for Cooperative and Scalable Feature Transformation

Researchers propose a new multi-agent reinforcement learning framework that uses three cooperative agents with attention mechanisms to automate feature transformation for machine learning models. The approach addresses key limitations in existing automated feature engineering methods, including dynamic feature expansion instability and insufficient agent cooperation.

• →New heterogeneous multi-agent RL framework enables more efficient automated feature transformation for structured data.
• →Framework uses three specialized agents with shared critic mechanism to improve cooperation and communication.
• →Multi-head attention mechanisms help agents handle dynamically expanding feature spaces during transformation.
• →State encoding technique stabilizes RL agent learning dynamics for more robust transformation policies.
• →Extensive experiments validate the model's effectiveness, efficiency, robustness, and interpretability.