Linear Adversarial Concept Erasure
Shauli Ravfogel, Michael Twiton, Yoav Goldberg, Ryan Cotterell
TL;DR
We address the problem of removing a linear concept from fixed representations by identifying a bias subspace and projecting away from it. The core approach frames this as a maximin game over an orthogonal projection $P$ in $\mathcal{P}_K$ and predictor parameters, balancing bias erasure with information preservation. The paper provides closed-form solutions for linear regression and partial least squares, and introduces RLACE, a convex relaxation for logistic regression, enabling practical optimization. Empirically, the method demonstrates effective gender-bias removal in both static and contextualized representations while preserving performance on nonlinear classifiers, yielding interpretable, tractable debiasing with broad applicability.
Abstract
Modern neural models trained on textual data rely on pre-trained representations that emerge without direct supervision. As these representations are increasingly being used in real-world applications, the inability to \emph{control} their content becomes an increasingly important problem. We formulate the problem of identifying and erasing a linear subspace that corresponds to a given concept, in order to prevent linear predictors from recovering the concept. We model this problem as a constrained, linear maximin game, and show that existing solutions are generally not optimal for this task. We derive a closed-form solution for certain objectives, and propose a convex relaxation, \method, that works well for others. When evaluated in the context of binary gender removal, the method recovers a low-dimensional subspace whose removal mitigates bias by intrinsic and extrinsic evaluation. We show that the method is highly expressive, effectively mitigating bias in deep nonlinear classifiers while maintaining tractability and interpretability.