Ant Group Showcases AI Research at NeurIPS 2019: Deep Exponential Family and Graph Neural Network Innovations

The annual NeurIPS 2019 conference was held in Vancouver from December 8 to 14, attracting 6,743 valid paper submissions with an acceptance rate of 21.17%.

Ant Group (formerly Ant Financial) had multiple papers accepted, focusing on topics such as generative adversarial networks, graph neural networks, and deep reinforcement learning.

One highlighted paper, "Deep Exponential Family Distribution Estimation via Adversarial Dynamic System Embedding," addresses maximum‑likelihood estimation for the general deep exponential family by reformulating the intractable log‑partition function into an energy function and learning a negative sampler through a primal‑dual approach, enabling amortized inference.

Another highlighted work, "Inverse Synthesis Prediction with Conditional Graph Logic Networks," introduces a novel graph neural network that learns to apply chemical reaction templates for retrosynthesis analysis, achieving an 8.2% accuracy improvement over the previous state‑of‑the‑art while providing interpretable results.

Ant Group also demonstrated AI‑driven applications at the conference, including a smart waste‑sorting mini‑program and intelligent recycling bin that identify garbage types via image recognition.

Beyond NeurIPS, Ant Group regularly contributes to top AI conferences such as ICML, AAAI, SIGIR, and KDD, bridging cutting‑edge research with real‑world financial services.

The article concludes by announcing Ant Group’s new e‑book "Behind Double‑11: Alipay Technology Upgrade" and invites readers to follow the official "Ant Group Technology" public account for a free download.