H-SPLID: HSIC-based Saliency Preserving Latent Information Decomposition
Not on the map. Why?
Abstract (arXiv)
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.
Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Builds on 37 of its 65 references that are in the map. First posted on arXiv 23 Oct 2025.
Builds on
37 of 65 references.
References · 37
- 2015Deep Residual Learning for Image Recognitioncited by 52,096
- 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,305
- 2014Microsoft COCO: Common Objects in Contextcited by 17,608
- 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,733
- 2017A Unified Approach to Interpreting Model Predictionscited by 6,555
- 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,415
- 2017Axiomatic Attribution for Deep Networkscited by 3,384
- 2019Benchmarking Neural Network Robustness to Common Corruptions and Perturbationscited by 2,877
- 2013Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Mapscited by 2,371
- 2020Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attackscited by 1,451
- 2015Learning Transferable Features with Deep Adaptation Networkscited by 1,329
- 2023Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecturecited by 1,039
25 more
- 2016Deep Variational Information Bottleneckcited by 930
- 2018Robustness May Be at Odds with Accuracycited by 880
- 2015Deep Learning and the Information Bottleneck Principlecited by 802
- 2019Score-CAM: Score-Weighted Visual Explanations for Convolutional Neural Networkscited by 582
- 2017Skin Lesion Analysis Toward Melanoma Detection: A Challenge at the 2017 International Symposium on Biomedical Imaging (ISBI), Hosted by the International Skin Imaging Collaboration (ISIC)cited by 549
- 2020Noise or Signal: The Role of Image Backgrounds in Object Recognitioncited by 343
- 2017Improving the Adversarial Robustness and Interpretability of Deep Neural Networks by Regularizing their Input Gradientscited by 236
- 2020Torchattacks: A PyTorch Repository for Adversarial Attackscited by 158
- 2020SaliencyMix: A Saliency Guided Data Augmentation Strategy for Better Regularizationcited by 149
- 2019On the Connection Between Adversarial Robustness and Saliency Map Interpretabilitycited by 80
- 2019The HSIC Bottleneck: Deep Learning without Back-Propagationcited by 77
- 2020The Conditional Entropy Bottleneckcited by 75
- 2021Improving Deep Learning Interpretability by Saliency Guided Trainingcited by 68
- 2024A Sober Look at the Robustness of CLIPs to Spurious Featurescited by 38
- 2019Robust Learning with the Hilbert-Schmidt Independence Criterioncited by 36
- 2019Contrastive Variational Autoencoder Enhances Salient Featurescited by 28
- 2021Revisiting Hilbert-Schmidt Information Bottleneck for Adversarial Robustnesscited by 20
- 2022Moment Matching Deep Contrastive Latent Variable Modelscited by 13
- 2023Adversarial Examples Exist in Two-Layer ReLU Networks for Low Dimensional Linear Subspacescited by 8
- 2024Exploring the Interplay of Interpretability and Robustness in Deep Neural Networks: A Saliency-guided Approachcited by 7
- 2018Feature Prioritization and Regularization Improve Standard Accuracy and Adversarial Robustnesscited by 5
- 2024Effect of Ambient-Intrinsic Dimension Gap on Adversarial Vulnerabilitycited by 5
- 2024Data Augmentation via Latent Diffusion for Saliency Predictioncited by 5
- 2019Efficient Bayesian shape-restricted function estimation with constrained Gaussian process priorscited by 3
- 2023Saliency Guided Contrastive Learning on Scene Imagescited by 2
Built on by
Citing papers per year: 2026: 2.
2 papers.