analiza aiqbench
BiasFlow introduces a diagnostic toolkit and regularization method designed to mitigate spurious feature reliance in AI models by monitoring geometric properties like class-attribute centroid alignment. The approach demonstrates improved worst-group accuracy on specific benchmarks while reducing probe accuracy for biased attributes, offering a practical mechanism for evaluating model robustness against retraining on skewed data. This research advances the field of machine learning fairness by providing concrete metrics and penalties to ensure backbone representations remain resilient when new heads are trained.
BiasFlow: Geometric Monitoring and Backbone Regularization for Spurious Feature Reliance
arXiv:2610.06846v1 Announce Type: new Abstract: Worst-group accuracy (WGA) evaluates a trained predictor but does not characterize how its frozen backbone behaves when a new head is learned. We introduce BiasFlow, a hook-based toolkit for monitoring class-attribute centroid alignment (IBMI), within-class centroid separation (W-IBMI), and feature-projection sensitivity. IBMI is confounded by class-attribute correlation and is not a measure of causal feature reliance. We pair these diagnostics with BiasFlow Regularization (BFR), a supervised, composable class-conditional centroid-alignment penalty. W-IBMI verifies the quantity BFR optimizes; it is scale dependent and does not independently establish attribute removal. Across the reported small-scale benchmarks, adding BFR improves or preserves mean WGA, with gains up to +26.0 pp on UrbanCars. The principal independent stress test freezes CelebA-Std backbones and trains fresh heads on biased data: BFR+GroupDRO improves WGA from 40.7% to 64.1%, while Male probe accuracy decreases from 92.5% to 72.2%. Attribute information remains recoverable, and cross-task results are mixed. A controlled synthetic-watermark ImageNet experiment additionally improves watermark-shift accuracy by +23.0 pp under matched training. These results support evaluating centroid geometry and resistance to biased head retraining alongside WGA, within the tested protocols.
Abstract:Worst-group accuracy (WGA) evaluates a trained predictor but does not characterize how its frozen backbone behaves when a new head is learned. We introduce BiasFlow, a hook-based toolkit for monitoring class-attribute centroid alignment (IBMI), within-class centroid separation (W-IBMI), and feature-projection sensitivity. IBMI is confounded by class-attribute correlation and is not a measure of causal feature reliance. We pair these diagnostics with BiasFlow Regularization (BFR), a supervised, composable class-conditional centroid-alignment penalty. W-IBMI verifies the quantity BFR optimizes; it is scale dependent and does not independently establish attribute removal. Across the reported small-scale benchmarks, adding BFR improves or preserves mean WGA, with gains up to +26.0 pp on UrbanCars. The principal independent stress test freezes CelebA-Std backbones and trains fresh heads on biased data: BFR+GroupDRO improves WGA from 40.7% to 64.1%, while Male probe accuracy decreases from 92.5% to 72.2%. Attribute information remains recoverable, and cross-task results are mixed. A controlled synthetic-watermark ImageNet experiment additionally improves watermark-shift accuracy by +23.0 pp under matched training. These results support evaluating centroid geometry and resistance to biased head retraining alongside WGA, within the tested protocols.
Artificial Intelligence (cs.AI)
arXiv:2610.06846 [cs.AI]
(or arXiv:2610.06846v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.06846 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 274 cuvinte
Subiecte conexe
Energy-Regularized Spatial Masking: A Novel Approach to Enhancing Robustness and Interpretability in Vision Models
arXiv · 10 iun. 2026
Energy-Regularized Spatial Masking: A Novel Approach to Enhancing Robustness and Interpretability in Vision Models
arXiv · 9 iun. 2026
Prevalence calibration as shortcut mitigation
arXiv · 9 sept. 2026
[AINews] Meta Connect 2026: Muse glasses, voice, video, and Charm
Latent Space · 24 sept. 2026