H-SPLID: HSIC-based Saliency Preserving Latent Information Decomposition
This addresses robustness in feature learning for image classification, though it appears incremental as it builds on existing decomposition and HSIC methods.
The paper tackles the problem of learning salient feature representations by decomposing salient and non-salient features into separate spaces, showing that H-SPLID promotes low-dimensional, task-relevant features and reduces sensitivity to non-salient perturbations in image classification tasks.
We introduce H-SPLID, a novel algorithm for learning salient feature representations through the explicit decomposition of salient and non-salient features into separate spaces. We show that H-SPLID promotes learning low-dimensional, task-relevant features. We prove that the expected prediction deviation under input perturbations is upper-bounded by the dimension of the salient subspace and the Hilbert-Schmidt Independence Criterion (HSIC) between inputs and representations. This establishes a link between robustness and latent representation compression in terms of the dimensionality and information preserved. Empirical evaluations on image classification tasks show that models trained with H-SPLID primarily rely on salient input components, as indicated by reduced sensitivity to perturbations affecting non-salient features, such as image backgrounds. Our code is available at https://github.com/neu-spiral/H-SPLID.