The findings suggest that, in the studied setting, representation-level forgetting is primarily governed by gradient concentration and representation geometry rather than by the specific identity of saliency-selected weights.
Abstract
Machine unlearning aims to remove the influence of specific training data while preserving model utility. Many state-of-the-art approaches pursue this goal by restricting the forgetting update to a subset of parameters selected through gradient-based saliency. Although such methods are widely adopted, the actual contribution of saliency-based weight selection to representation-level forgetting remains unclear. In this work, we perform the first controlled ablation of the saliency masking mechanism used by SalUn. Using a matched-compute experimental design on CIFAR-10 and CIFAR-100 with ResNet-18, we compare saliency-based masking against random masks of equal sparsity and unconstrained updates, while keeping the unlearning objective, optimization schedule, and computational budget fixed. Across multiple representation-level evaluations, including linear probing, prototype recovery, and layer-wise CKA, the three configurations exhibit statistically equivalent representation-level recoverability. We find that forget gradients are strongly concentrated in the final network layers (approximately 92% of the squared gradient energy on CIFAR-10) before any mask is applied, causing all masking strategies to operate within the same representational subspace. Furthermore, saliency masks show limited class specificity (specificity index 0.09-0.11), selecting highly overlapping parameter subsets across different forget classes. Our findings suggest that, in the studied setting, representation-level forgetting is primarily governed by gradient concentration and representation geometry rather than by the specific identity of saliency-selected weights. More broadly, the results support a growing body of evidence indicating that effective representation-level unlearning requires objectives that act directly on latent representations rather than on increasingly sophisticated weight-selection strategies.
Machine unlearning (MU) aims to remove the influence of specific training data while preserving model utility. As the name suggests, MU can be viewed as the inverse of learning, using gradient-based updates to reduce the influence of a forget-set by counteracting the previously learned behavior. Recently, Muon, a gradient descent variant, has been introduced. Muon applies spectral magnitude normalization to encourage exploration of rare directions and demonstrates promising performance. Inspired by Muon, we adopt the spectral view for unlearning and propose Spectral Saliency Unlearning (SSU). SSU thresholds weak singular components and updates only those directions supported by a confident unlearning signal. We further provide theoretical justification for this thresholding approach from the perspective of the forgetting-retention trade-off. Experiments across image classifiers, diffusion models, and LLMs demonstrate SSU's effectiveness.
Cedar Site Bai, Amber Yijia Zheng, Raymond A. Yeh et al.· 0 citations
Biological vision has evolved to make efficient use of the limited information processing capability and tight energy budget of the brain by preferentially processing the most salient features of visual scenes. In contrast, modern deep vision models rely on expansive, high-dimensional representations. This may offer potential recognition gains but increases computing costs. As a consequence, the computer vision community has been exploring sparsity-enforcing techniques such as activation dropout and weight pruning. Beyond ameliorating the burden of computation, sparsity techniques such as random dropout have been shown to regularize model training, thus allowing for better generalization. Here, we pursue both dropout and weight pruning in tandem and adopt a biologically plausible saliency-informed dropout technique as an explainable alternative to the unstructured standard random dropout approach. Our approach involves a hierarchical “retinotopic" gating of convolutional feature maps, which promotes efficient deletion of “redundant" weights by iterative magnitude pruning. We explore the effectiveness of saliency-informed dropout based on different approaches to dropping based on the saliency map, dropout through all layers or only early layers, and by additionally applying dropout at inference. We compare throughout with standard random dropout. We present empirical results on regimes where such structured dynamic and static (weight) sparsities interact optimally to prune a ResNet model on the Imagenette dataset.
Shira Goldhaber-Gordon, A. Akwaboah, Aaron L. Sampson et al.· International Conference on...· 1 citation
Machine unlearning (MU) aims to remove the influence of selected data from trained models, offering an efficient alternative to full retraining. With the rise of increasingly stringent privacy regulations, including the right to be forgotten, machine learning models must incorporate mechanisms that ensure compliance while preserving performance. However, existing MU methods often suffer from performance degradation, and limited granularity, leaving residual traces that adversaries can exploit. We propose DeepU, a fine-grained unlearning framework that performs within-layer, weight-level analysis and quantifies the influence of individual weights using the signal-to-noise ratio (SNR) derived from the gradients of sensitive and non-sensitive data. DeepU categorizes weights as influential, intra-dependent, or non-influential and applies targeted weight updates: resetting or perturbing influential weights, decaying intra-dependent ones, and stabilizing the remaining ones via limited backpropagation. We evaluated DeepU on standard datasets including CIFAR-10, CIFAR-100, Tiny ImageNet, and CelebA, demonstrating a 60–90% reduction in successful membership inference attacks with less than a 3% drop in accuracy, significantly outperforming existing benchmark methods. Re-tuning requires only 20.75 seconds and 102.47 MB, making DeepU up to 36.6 times faster than competing methods. We further evaluate robustness against Generative Adversarial Network (GAN) and diffusion-based model inversion attacks across white-box, black-box, and label-only threat models, demonstrating that DeepU consistently reduces reconstruction of forgotten data.
This work proposes a novel one-shot unlearning approach, abandoning iterative optimization in favor of a direct, exact analytical solution, and achieves state-of-the-art forgetting-utility trade-offs on TOFU-5%, TOFU-10%, MUSE-Books, MUSE-News and WMDP, significantly reducing computational overhead without sacrificing overall model performance.
Paweł Batorski, P. Spurek, Paul Swoboda· 0 citations
This paper studies what happens to the rest of the model when a class is forgotten, using a label-conditioned energy-based model (EBM) that assigns per-class energies, making the effect directly observable.
Syed Ali Ahmed, Syed Bilal Ahsan, Muhammad Zaigham Zaheer National University of Computer et al.· 0 citations
Dataset distillation aims to synthesize compact datasets that can approximate the performance of full-data training while significantly reducing computational and storage costs. However, diffusion-based distillation methods often struggle to preserve structural coherence and generalization, especially in visually complex domains. This issue often stems from latent prototypes that are weakly aligned with class-discriminative regions and contaminated by irrelevant background, thereby degrading generation quality and generalization. To address this limitation, we propose a saliency-driven distillation framework that constructs class-discriminative latent prototypes to enhance representativeness and generalization. The framework proceeds in two stages: (1) ensemble Grad-CAM++ saliency is used to construct prototypes emphasizing class-discriminative regions, and (2) hard-prototype refinement is then applied to construct challenging yet class-consistent prototypes, thereby enhancing discriminability and diversity. Importantly, the diffusion backbones (e.g., LDM and DiT) remain frozen; only lightweight classifiers used for saliency extraction are trained. Extensive experiments across multiple benchmarks demonstrate consistent performance improvements over strong baselines. Code will be released.
Yawen Zou, Wenqi Cai, Guang Li et al.· arXiv.org· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
