Excited to share that two of our Neural Graph Database papers have been accepted this season.

1. NGDB-Zoo: Towards Efficient and Scalable Neural Graph Databases Training

   • Authors: Zhongwei Xie, Jiaxin Bai, Shujie Liu, Haoyu Huang, Yufei Li, Yisen Gao, Hong Ting Tsang, Yangqiu Song
   • Accepted to NeurIPS 2026. NGDB-Zoo tackles training bottlenecks in Neural Graph Databases by combining operator-level training with semantic augmentation—decoupling logical operators from query topologies for multi-stream parallelism, and integrating Pre-trained Text Encoder priors without I/O or memory blow-ups. Code: HKUST-KnowComp/neural_graph_db_training.

2. NGDBench: Towards Neural Graph Data Management

   • Authors: Yufei Li, Yisen Gao, Jiaxuan Xiong, Jiaxin Bai, Shijie Zhong, Haoyu Huang, Zhongwei Xie, Hong Ting Tsang, Yangqiu Song
   • Accepted to EMNLP 2026 Findings. NGDBench is a multi-domain benchmark for neural graph data management: each instance pairs a clean latent graph with a realistically perturbed observed graph, supports full Cypher and dynamic operations, and stresses Text-to-Cypher and GraphRAG under noise and continuous updates. Code: HKUST-KnowComp/NGDBench.

Huge congratulations to Zhongwei and Yufei for the hard work that carried these projects through—looking forward to sharing more with the community at NeurIPS and EMNLP.