Isoact improves robust representations by respecting manifold geometry
IsoAct: Geometry-Aware Post-hoc Debiasing via Isometric Actions
Machine Learning
Summary
Designing computer programs that understand data well often means ignoring unrelated details called nuisances. This is tricky when data lies on curved or complex shapes rather than flat spaces. The authors created IsoAct, a method that cleverly finds and fixes these nuisance directions while keeping the data on its natural curved shape. Their tests show IsoAct reduces unwanted influences without hurting important information or breaking the data’s structure.
What this means in practice
- •For machine learning engineers: Improve model robustness by removing nuisance information from learned features that live on complex geometric spaces.
- •For robotics developers: Enhance pose representations in robotics by debiasing rigid body transformations without distorting geometric validity.
Authors
Sumin Park, Taero Kim, Minhyeong Cho, Subeen Park, Jong Hwan Lee, Kyungwoo Song
Abstract
Learning representations that are robust to nuisance factors is essential for reliable deployment, but remains challenging when the representations lie on non-Euclidean manifolds. Existing debiasing methods typically assume Euclidean representations or apply linear interventions, which can distort the geometry of structured latent spaces. We propose IsoAct, a post-hoc framework that accounts for geometry in both nuisance identification and representation editing. It identifies nuisance-predictive directions using Fisher discriminant analysis (FDA) in tangent coordinates and applies corrections through manifold-native actions that preserve manifold membership. Controlled and real-world experiments on hyperspherical, hyperbolic, and $\mathrm{SE}(3)$ representations evaluate nuisance reduction, utility preservation, and manifold validity jointly. IsoAct reduces nuisance recoverability while retaining competitive utility. Its clearest joint result is on KLIFS, combining reduced kinase-group recoverability, strong structural prediction, and valid rigid poses.