FAROS: Fair Graph Generation via Attribute Switching Mechanisms
Abdennacer Badaoui, Oussama Kharouiche, Hatim Mrabet, Fragkiskos D. Malliaros · 2025
Read FAROS: Fair Graph Generation via Attribute Switching Mechanisms by Abdennacer Badaoui, Oussama Kharouiche, Hatim Mrabet, Fragkiskos D. Malliaros (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-2507-03728.txt.
Description
Abdennacer Badaoui and co-authors generate fair synthetic networks without retraining the generator. Their 2025 paper tackles graph diffusion models, which synthesize realistic network structures but can bake in bias tied to traits like group membership. Retraining such models with fairness constraints is expensive and clumsy.
Their framework, FAROS, intervenes mid-generation instead: it flips the sensitive trait of a carefully chosen fraction of nodes at a carefully chosen diffusion step. The fraction and timing come from balancing two proxies, keeping the network's structural profile close to the original while making links independent of the sensitive trait. On link prediction benchmarks, the method cut fairness gaps while matching or beating rivals' accuracy. In some settings it achieved the best trade-off outright. The idea generalizes beyond graphs: fairness as an edit during generation, applied to a frozen model, is cheaper than fairness as a rebuild.
Catalog record
| CREATOR | Abdennacer Badaoui, Oussama Kharouiche, Hatim Mrabet, Fragkiskos D. Malliaros |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 72 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-03728