Ant Group Open-sources Ant Graph Learning (AGL), the First General Industrial Graph Learning System

On the afternoon of September 7, at the 2023 Inclusion·Bund Conference in Shanghai, Ant Group officially open‑sourced Ant Graph Learning (AGL), the industry's first general industrial graph learning system.

AGL currently enables information collaboration and structural awareness on trillion‑scale graph data, has built multiple industry digital graph intelligence solutions, and contributed over 60 CCF‑A/B journal and conference papers, 40+ patents, and participated as a core contributor to the national graph neural network standard.

The open‑source AGL v0.1 release provides Ant Group's mature industrial‑grade graph learning system together with a suite of ready‑to‑use graph learning algorithms, with the code repository launched on GitHub on the same day.

Through open‑sourcing, AGL offers a full‑stack solution for large‑scale industrial graph learning tasks, aiming to empower developers with a powerful platform, lower the entry barrier for graph learning, foster community‑driven innovation, and promote widespread adoption of graph AI across sectors.

Since 2017, graph learning has become a hot research topic in artificial intelligence and a strategic focus for global technology competition; China’s "14th Five‑Year" plan emphasizes breakthroughs in large‑scale parallel graph processing, heterogeneous data management, and new machine‑learning technologies.

AGL has been widely applied within Ant Group’s diversified businesses, delivering notable results such as a 50% increase in supply‑chain identification accuracy and raising loan approval rates from 30% to 80% in the Net Business Bank digital supply‑chain “Great Goose” system, as well as improving supply distribution efficiency by over 50% on the Alipay digital open platform.

At the forum, experts from Tsinghua University, Zhejiang University, and Sun Yat‑sen University discussed data‑driven intelligent decision making, diffusion‑model approaches for offline reinforcement learning, and robust optimization methods for digital‑economy finance.