Graph Neural Networks with Similarity-Navigated Probabilistic Feature Copying Journal: Nuclear Physics B
Asela Hevapthige · 2025
Read Graph Neural Networks with Similarity-Navigated Probabilistic Feature Copying Journal: Nuclear Physics B by Asela Hevapthige (2025) 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-2509-19084.txt.
Description
Asela Hevapathige, writing alone, designs a graph neural network meant to avoid three known weaknesses of the breed. Graph networks pass messages between linked points to classify them, but deep versions blur all points into look-alikes, a flaw called oversmoothing. They also handle mixed link types poorly and lump all features together in one block. The new design, AxelGNN, borrows from a model of how culture spreads between similar people.
Its links open or close based on how similar two points are, letting groups converge or drift apart. Features are copied in segments, small meaningful groups at a time, rather than all at once. A global polarization step keeps several distinct clusters alive, fighting oversmoothing. Tests show the design matches or beats prior methods on node classification and influence estimation, on both friendly and mixed-link graphs.
Catalog record
| CREATOR | Asela Hevapthige |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-19084