# Gradient Concentration, Not Weight Saliency, Explains Representation-Level Class Unlearning

arXiv: https://arxiv.org/abs/2607.21353 · HTML: https://mlsanity.com/paper/2607.21353/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Machine unlearning, one of 12 directions in LLM safety and security.
- Builds on 18 of its 57 references that are in the map.
- First posted on arXiv 23 Jul 2026.

## Placed in

- LLM safety and security / [Machine unlearning](https://mlsanity.com/safety/machine-unlearning/)

## Builds on (most cited first, 18 of 18)

- [Making AI Forget You: Data Deletion in Machine Learning](https://mlsanity.com/paper/1907.05012/) (2019)
- [SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation](https://mlsanity.com/paper/2310.12508/) (2023)
- [A Survey of Machine Unlearning](https://mlsanity.com/paper/2209.02299/) (2022)
- [Inexact Unlearning Needs More Careful Evaluations to Avoid a False Sense of Privacy](https://mlsanity.com/paper/2403.01218/) (2024)
- [Machine Unlearning Fails to Remove Data Poisoning Attacks](https://mlsanity.com/paper/2406.17216/) (2024)
- [A Comprehensive Survey of Machine Unlearning Techniques for Large Language Models](https://mlsanity.com/paper/2503.01854/) (2025)
- [Corrective Machine Unlearning](https://mlsanity.com/paper/2402.14015/) (2024)
- [Are We Truly Forgetting? A Critical Re-examination of Machine Unlearning Evaluation Protocols](https://mlsanity.com/paper/2503.06991/) (2025)
- [Gone but Not Forgotten: Improved Benchmarks for Machine Unlearning](https://mlsanity.com/paper/2405.19211/) (2024)
- [Contrastive Representation Shaping for LLM Unlearning](https://mlsanity.com/paper/2601.22028/) (2026)
- [POUR: A Provably Optimal Method for Unlearning Representations via Neural Collapse](https://mlsanity.com/paper/2511.19339/) (2025)
- [Erase at the Core: Representation Unlearning for Machine Unlearning](https://mlsanity.com/paper/2602.05375/) (2026)
- [Unlearning's Blind Spots: Over-Unlearning and Prototypical Relearning Attack](https://mlsanity.com/paper/2506.01318/) (2025)
- [An Illusion of Unlearning? Assessing Machine Unlearning Through Internal Representations](https://mlsanity.com/paper/2604.08271/) (2026)
- [Class Unlearning via Depth-Aware Removal of Forget-Specific Directions](https://mlsanity.com/paper/2604.15166/) (2026)
- [One-Point Contraction: Erasing Representational Separability toward Irreversible Deep Forgetting](https://mlsanity.com/paper/2507.07754/) (2025)
- [Representation Unlearning: Forgetting through Information Compression](https://mlsanity.com/paper/2601.21564/) (2026)
- [Erased, but Not Gone: Output Forgetting Is Not True Forgetting](https://mlsanity.com/paper/2606.25001/) (2026)

## Abstract (arXiv)

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.