Skip to yearly menu bar Skip to main content


Poster
in
Workshop: CODEML: Championing Open-source DEvelopment in Machine Learning

OpenDG: A Modular Framework for Machine Learning on Dynamic Graphs

Jacob Chmura · Shenyang (Andy) Huang · Ali Parviz · Farimah Poursafaei · Michael Bronstein · Guillaume Rabusseau · Matthias Fey · Reihaneh Rabbany

[ ] [ Project Page ]
Fri 18 Jul 2:15 p.m. PDT — 3 p.m. PDT

Abstract:

While deep learning on static graphs has been revolutionized by standardized libraries like PyTorch Geometric and DGL, machine learning on dynamic graphs, networks that evolve over time, lacks comparable software infrastructure. Existing libraries are limited in scope, focusing on narrow method categories or specific algorithms. We introduce OpenDG, a comprehensive framework for machine learning on dynamic graphs that bridges this gap. OpenDG provides validated implementations of widely used temporal graph learning methods, supporting both discrete and continuous-time approaches through a modular architecture. The framework’s decoupled design combines an intuitive front-end API with an optimized backend, enabling reproducible research and efficient experimentation at scale. Key features include graph-level optimizations for offline training and built-in performance profiling capabilities. Through extensive benchmarking on five real-world networks, OpenDG is up to 6 times faster than the widely used DyGLib library on TGN and TGAT models and up to 8 times faster than the UTG framework for converting edges into coarse-grained snapshots.

Chat is not available.