Reproduce “Hydra: Towards Transferable Multi-Task Learning on Temporal Graphs” 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

Kiarash Shamsi, Farimah Poursafaei, Tran Gia Bao Ngo, Reihaneh Rabbany, Baris Coskunuzer, Guillaume Rabusseau, Michael M. Bronstein, Shenyang Huang and 1 more

Hydra reuses a model pretrained across source networks to predict multiple graph-wide properties on unseen networks without target-specific training or fine-tuning. It combines global spectral summaries with temporal modeling of local graph structure and shares the representation across task-specific heads. On held-out networks, Hydra-32 achieves mean ROC-AUC 0.759 for largest-connected-component growth/shrinkage, versus 0.686 for GraphPulse; GraphPulse is trained on test networks.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.