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

TL;DR

NGDB-Zoo introduces a unified framework for neural graph database training that enhances efficiency and scalability by decoupling logical operators and integrating semantic priors, leading to significant throughput improvements.

Contribution

It proposes a novel decoupled architecture and dynamic data-flow execution for neural graph databases, enabling multi-stream parallelism and semantic augmentation without memory issues.

Findings

Achieves 1.8x to 6.8x throughput improvements over baselines.

Maintains high GPU utilization across diverse logical patterns.

Effectively integrates semantic priors from pre-trained text encoders.

Abstract

Neural Graph Databases (NGDBs) facilitate complex logical reasoning over incomplete knowledge structures, yet their training efficiency and expressivity are constrained by rigid query-level batching and structure-exclusive embeddings. We present NGDB-Zoo, a unified framework that resolves these bottlenecks by synergizing operator-level training with semantic augmentation. By decoupling logical operators from query topologies, NGDB-Zoo transforms the training loop into a dynamically scheduled data-flow execution, enabling multi-stream parallelism and achieving a $1.8\times$ - $6.8\times$ throughput compared to baselines. Furthermore, we formalize a decoupled architecture to integrate high-dimensional semantic priors from Pre-trained Text Encoders (PTEs) without triggering I/O stalls or memory overflows. Extensive evaluations on six benchmarks, including massive graphs like ogbl-wikikg2 and ATLAS-Wiki, demonstrate that NGDB-Zoo maintains high GPU utilization across diverse logical patterns and significantly mitigates representation friction in hybrid neuro-symbolic reasoning.