Adversarial Attacks on Hyperbolic Networks
Max van Spengler, Jan Zahálka, Pascal Mettes · 2024
Read Adversarial Attacks on Hyperbolic Networks by Max van Spengler, Jan Zahálka, Pascal Mettes (2024) 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-2412-01495.txt.
Description
Max van Spengler, Jan Zahálka, and Pascal Mettes present a 2024 paper on fooling neural networks that live in curved space. Hyperbolic networks embed data in a saddle-shaped geometry that suits hierarchies, and they are gaining users. Nobody had checked how easily tiny, crafted perturbations can flip their answers.
The authors port two standard attacks into hyperbolic geometry and build clear synthetic benchmarks beside real data tests. Old and new attacks behaved differently across the settings. More telling, hyperbolic and ordinary networks failed in different ways, and the new attacks could not erase that gap. The authors conclude the robustness shift comes from the geometry teaching the model different patterns, not from attack weakness. Defenders of curved models need geometry-aware shields.
Catalog record
| CREATOR | Max van Spengler, Jan Zahálka, Pascal Mettes |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-01495