Reproduce “AEGNN: Asynchronous Event-based Graph Neural Networks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Simon Schaefer, Daniel Gehrig, Davide Scaramuzza
AEGNN processes event camera data using asynchronous, evolving graph neural networks, demonstrating orders-of-magnitude reduction in computational load (e.g., 3000-fold FLOPs reduction compared to dense methods) and lower latency while maintaining or improving performance on object recognition and detection.
Key insights
- Information propagation in Graph Neural Networks can be localized to affected k-hop subgraphs, enabling efficient, partial recomputation when new data arrives.
- Deep learning models can be trained synchronously on batch event data and then seamlessly converted for highly efficient asynchronous, event-by-event deployment.
- Representing events as nodes in an evolving spatio-temporal graph allows for direct processing of sparse event data, preserving its intrinsic properties.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.