AlphaMemo: Structured Search-Process Memory for Self-Evolving Alpha Mining Agents

AlphaMemo is introduced as a self-evolving alpha mining agent that utilizes Structured Search-Process Memory to enhance the efficiency and effectiveness of financial model development. It innovatively records reusable evidence of successful and unsuccessful edit motifs derived from Abstract Syntax Tree (AST) differences, employing techniques like confidence-gated residual memory and asymmetric veto control to mitigate overfitting and redundancy. Experimental results on the CSI 500 and S&P 500 datasets demonstrate improved out-of-sample performance and discovery efficiency, making it a valuable tool for practitioners in financial AI modeling.