Ant Group Open-Sources OpenASCE: A Distributed Full-Stack Causal Learning System Presented at NeurIPS

On December 10, the premier machine learning and artificial intelligence conference NeurIPS opened in New Orleans, gathering AI experts from industry and academia worldwide.

During the first day, Ant Group announced the open‑source release of OpenASCE (Open All‑Scale Causal Engine) at a session titled “Knowledge‑Enhanced AI in Vertical Industries”.

Causal inference studies how to infer cause‑effect relationships from data, complementing traditional machine learning which relies on correlations; combining the two yields causal learning, a technique that deepens understanding of data and decision‑making processes.

OpenASCE builds on Ant Group’s years of practice and breakthroughs, offering a full‑link, large‑scale causal learning framework that covers causal discovery, effect estimation, and attribution across all causal domains.

For causal discovery, OpenASCE supports distributed Bayesian network structure search capable of handling hundreds of nodes with millions of samples, and continuous‑optimization‑based discovery that scales to tens of thousands of nodes with billions of samples.

The system’s large‑scale distributed causal uplift tree can train 100 million samples within four hours, making it the only distributed causal uplift tree implementation in the industry.

OpenASCE also incorporates more than 20 industrial‑grade causal algorithms, including over 15 methods that integrate causal techniques with deep learning, lowering the barrier for industrial adoption and enabling large‑scale applications within Ant Group.

In credit risk control, OpenASCE’s causal learning more accurately identifies causal links between risk factors and customer behavior, greatly improving precision and efficiency; in marketing optimization, it helps locate “marketing‑sensitive” audiences; in recommendation, causal inference corrects data bias, removes spurious correlations, and learns more stable causal relationships.

By open‑sourcing OpenASCE, Ant Group provides the community with a high‑performance, large‑scale causal learning solution and invites global developers to contribute, fostering innovation in full‑link causal learning systems.

Open‑source is a core technical strategy for Ant Group, aiming to build an open, inclusive ecosystem that shares technological benefits widely.

To date, Ant Group has open‑sourced over 1,700 repositories in databases, cloud‑native, middleware, and other foundational software areas, ranking among the top three in China’s open‑source influence rankings, with nine major technologies supporting the core of Alipay.