Kernelized Edge Attention: Addressing Semantic Attention Blurring in Temporal Graph Neural Networks
Govind Waghmare, Srini Rohan Gujulla Leel, Nikhil Tumbde, Sumedh B G, Sonia Gupta, Srikanta Bedathur · 2026
Read Kernelized Edge Attention: Addressing Semantic Attention Blurring in Temporal Graph Neural Networks by Govind Waghmare, Srini Rohan Gujulla Leel, Nikhil Tumbde, Sumedh B G, Sonia Gupta, Srikanta Bedathur (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-00596.txt.
Description
Govind Waghmare and colleagues address attention blurring in temporal graph neural networks, where attention over mixed node and edge states misses their temporal behaviour. Node vectors drift with long term structure, while edge data capture timed interactions such as messages, trades, or transactions. Their KEAT method tunes edge features with continuous-time kernels, including learnable MLP, Laplacian, and RBF variants, keeping node and edge roles apart. It fits Transformer and message passing designs, beating DyGFormer by up to 18% and TGN by 7% on link prediction MRR.
Catalog record
| CREATOR | Govind Waghmare, Srini Rohan Gujulla Leel, Nikhil Tumbde, Sumedh B G, Sonia Gupta, Srikanta Bedathur |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 79480 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602.00596