FLASH: Flexible Learning of Adaptive Sampling from History in Temporal Graph Neural Networks
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909573559877632 |
|---|---|
| author | Feldman, Or Mantri, Krishna Sri Ipsit Schönlieb, Carola-Bibiane Baskin, Chaim Eliasof, Moshe |
| author_facet | Feldman, Or Mantri, Krishna Sri Ipsit Schönlieb, Carola-Bibiane Baskin, Chaim Eliasof, Moshe |
| contents | Aggregating temporal signals from historic interactions is a key step in future link prediction on dynamic graphs. However, incorporating long histories is resource-intensive. Hence, temporal graph neural networks (TGNNs) often rely on historical neighbors sampling heuristics such as uniform sampling or recent neighbors selection. These heuristics are static and fail to adapt to the underlying graph structure. We introduce FLASH, a learnable and graph-adaptive neighborhood selection mechanism that generalizes existing heuristics. FLASH integrates seamlessly into TGNNs and is trained end-to-end using a self-supervised ranking loss. We provide theoretical evidence that commonly used heuristics hinders TGNNs performance, motivating our design. Extensive experiments across multiple benchmarks demonstrate consistent and significant performance improvements for TGNNs equipped with FLASH. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_07337 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | FLASH: Flexible Learning of Adaptive Sampling from History in Temporal Graph Neural Networks Feldman, Or Mantri, Krishna Sri Ipsit Schönlieb, Carola-Bibiane Baskin, Chaim Eliasof, Moshe Machine Learning Social and Information Networks Aggregating temporal signals from historic interactions is a key step in future link prediction on dynamic graphs. However, incorporating long histories is resource-intensive. Hence, temporal graph neural networks (TGNNs) often rely on historical neighbors sampling heuristics such as uniform sampling or recent neighbors selection. These heuristics are static and fail to adapt to the underlying graph structure. We introduce FLASH, a learnable and graph-adaptive neighborhood selection mechanism that generalizes existing heuristics. FLASH integrates seamlessly into TGNNs and is trained end-to-end using a self-supervised ranking loss. We provide theoretical evidence that commonly used heuristics hinders TGNNs performance, motivating our design. Extensive experiments across multiple benchmarks demonstrate consistent and significant performance improvements for TGNNs equipped with FLASH. |
| title | FLASH: Flexible Learning of Adaptive Sampling from History in Temporal Graph Neural Networks |
| topic | Machine Learning Social and Information Networks |
| url | https://arxiv.org/abs/2504.07337 |