Group 1 - The core conclusion highlights that Transformer enhances stock return prediction accuracy through spatiotemporal integration and multi-relation modeling, with GrifFinNet as a representative model [1][2] - Agent serves as a comprehensive decision-making entity in quantitative investment, simulating a professional investment process through a layered multi-agent framework, addressing challenges in traditional quantitative models [1][3] - The deep coupling of Transformer and Agent creates an integrated system that enhances both modeling precision and decision automation, facilitating a seamless transition from feature modeling to real trading [1][4] Group 2 - Transformer is identified as an efficient modeling architecture for quantitative investment, overcoming limitations of traditional models in handling nonlinear relationships and dynamic time series [2][12] - GrifFinNet, a key model based on Transformer, significantly outperforms traditional tools like LSTM and XGBoost in stock return prediction accuracy, demonstrating its effectiveness in the A-share market [2][24] - The Agent framework addresses issues in traditional quantitative investment by establishing a hierarchical structure that integrates macro selection, company analysis, portfolio optimization, and risk control [3][25] Group 3 - The integration of Transformer and Agent is not merely additive but follows a logic of functional complementarity, enhancing the overall efficiency of quantitative investment processes [4][28] - The multi-agent system designed for fundamental investing effectively combines structured and unstructured data, improving decision-making capabilities and adaptability to market changes [3][26] - Future advancements in AI-enabled quantitative investment will focus on precision, automation, and robustness, with ongoing optimization of both Transformer and Agent systems [4][33]