- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 1
Read review
Paper summary
This paper proposes a fine-grained, gradient-based unlearning that utilizes SVD for the unlearning objective gradient refinement.
The proposed method uses SVD to decompose a combined unlearning objective gradient (), and discards components correspond to small singular values. The hypothesizes is that those components are directions that have gradient conflict between the forget and retain objectives.
The paper then provides substantial amount of derivations to support their SVD-based gradient filtering.
The paper also provides experimental evidence over a diverse set of models: image classifiers, image diffusion models, and LLMs.
Critical strengths
-
The method is simple and concise. The proposed algorithm recipe is easy to follow and understand: main thing is to apply SVD decomposition over the computed unlearning gradients, and tune a singular value threshold for gradient refinement.
-
Experiment covers models for different tasks. The proposed method is demonstrated work for image classifier, image diffusion models, and LLMs.
Critical weaknesses
-
Lack of significant technical and engineering details. The paper delivers method and derivations utilizing unlearning objective , which breaks down to , but these objectives is never concretely written or defined.
- For example, is and actually in the same form but just the sign being different, or they are identical just the input data being different?
- This is important because the paper later on experiments their method on models of different tasks, and each of them has a significantly different training objectives.
- In Algorithm 1, the superscript appeared in line-4 but never introduced anywhere else. Does it correspond to layer of the networks in the actual implementation? If so, does the singular value threshold also needs to be tuned as -wise (i.e., different will need a different ), then how many layers did they consider? In this case the proposed method appeared to be too heuristic, which looks nice at high-level notations, but requires heavy engineering effort when it comes to the actual implementation (apparently it doesn't scale if you think of a large model that have many layers --- it is also non-transparent how you did that on LLMs).
-
Lack of basic experimental evidence. The key assumption the paper makes is that the singular-value structure of the sum of two gradients reveals the alignment or conflict between the individual gradient ( and ), yet the paper shows no direct evidence on that. Even though the paper currently provided pack of derivations and experiments, but it will be more convincing if the paper can have direct evidence from basic gradient analysis:
- Measure the forget and retain graidents separately in the combined SVD basis: for each , report and , check if holds at first hand. Then test whether low-singular-value components actually show more opposition than high-singular-value components.
- Back to the first bullet, the paper never defines and so it is also unclear how they can do such a gradient check.
[Minor]
- In Line 88-89, the notation for two gradient should specified what variable that the gradient is with respect to (reviewer assumes it is )
Questions for authors
-
Could you provide technical details, regarding how are and defined?
- Subquestion: how are and defined for each task model (image classifiers, image diffusions, LLMs)
-
Could you provide sufficient implementation details for better understanding of the experiment and its reproducibility, beyond the current Appendix B?
-
Could you provide basic gradient analysis and its implementation details, as specified in the second bullet of the Critical Weaknesses?
3 comments
Response to Reviewer 6tAr (2/3)
3. Actual implementation and layers considered (W1, Q2). For each task model, the layers we consider are fixed by layer type:
- Classification (ResNet-18): SSU filters the gradients of all 20 convolutions and the final linear layer with an exact SVD at every step. For the BatchNorm and bias parameters, which are vectors, SSU instead keeps the fraction of gradient coordinates with the largest magnitude.
- DDPM: SSU filters the 89 convolution kernels of the U-Net with an exact SVD at every step, and all other parameters use a coordinate mask computed once from the forget-set gradient.
- Stable Diffusion: SSU filters the 64 cross-attention matrices. Following SalUn, we compute the saliency scores once before unlearning as the entrywise magnitude of the forget-set gradient. At every step, we then project the gradient of each matrix onto the top singular directions of its saliency scores. All other parameters use a coordinate mask computed from the same scores.
- Llama-2-7B: SSU filters the 96 q/k/v projections of the 32 decoder layers with a randomized rank- SVD on every fifth micro-batch. All other parameters use a coordinate mask computed once from the forget-set gradient, which is also applied to q/k/v on the other micro-batches.
o address the concern that SSU may not scale to a large model with many layers, we timed the SVD routine of our training code for the full Llama-2-7B model on one A10G (median of repeated calls after warm-up):
| SVD of q/k/v (4096 × 4096) | Time ↓ | Peak extra GPU memory ↓ |
|---|---|---|
| Exact, one matrix | 3.86 s | 386 MiB |
| Randomized rank-819, one matrix | 62.6 ms | 333 MiB |
| Randomized, all 96 matrices | 6.0 s | 333 MiB |
| Amortized over five micro-batches | 1.2 s per micro-batch | – |
We observe that the randomized SVD is about 60× faster than an exact one, and since the matrices are filtered one at a time, the peak memory grows only by one matrix's workspace. We will include these discussion in Sec. 5.3 of the revised version.
Finally, for every task model the filtered gradient is passed to the optimizer of the base method (SGD with momentum, Adam or AdamW), and the plain gradient step in Algorithm 1 is a simplification for presentation. We will state this in Algorithm 1 and extend Appendix B with the matrix shapes, , step structure and optimizer settings of every task model. We will release the code upon acceptance.
4. Gradient notation (Minor). We sincerely thank the reviewer for pointing this out. The gradients are taken with respect to the weights, ; in practice per weight matrix on a mini-batch (likewise and ). We will correct this in the updated revision.
Response to Reviewer 6tAr (3/3)
5. Basic gradient analysis (W2, Q3). We thank the reviewer for this suggestion and have run the requested analysis. Specifically, on Tiny-ImageNet (three seeds) and DDPM (SSU on the convolution kernels or on all layers), we project the forget and retain gradients onto each singular direction of the combined gradient , i.e., and , and first verify that holds (up to floating-point error). We then compare the directions SSU keeps with those it truncates, before and after unlearning, by the share of opposed directions (), the mean alignment of Def. 4.1, and the first-order progress on each loss ( and , in units of ). The following table reports kept / truncated values; parentheses count the cases in which the truncated directions are more opposed:
| Setting | Opposed share | Alignment (Def. 4.1) | Progress on | Progress on |
|---|---|---|---|---|
| Tiny-ImageNet, start | 0.051 / 0.056 (3/3) | 0.117 / −0.011 | 0.936 / 0.0561 | 0.007 / 0.0009 |
| Tiny-ImageNet, end | 0.000 / 0.007 (3/3) | 0.046 / −0.293 | 0.398 / 0.0278 | 0.526 / 0.0475 |
| DDPM conv-only, start | 0.025 / 0.061 (3/3) | −0.085 / −0.431 | 0.102 / 0.0124 | 0.759 / 0.1267 |
| DDPM conv-only, end | 0.120 / 0.124 (7/9) | −0.033 / −0.326 | 0.004 / 0.0005 | 0.863 / 0.1331 |
| DDPM all-layers, start | 0.025 / 0.058 (3/3) | −0.173 / −0.459 | 0.105 / 0.0122 | 0.758 / 0.1243 |
| DDPM all-layers, end | 0.102 / 0.115 (7/7) | −0.116 / −0.352 | 0.004 / 0.0005 | 0.864 / 0.1313 |
We observe that:
- The truncated directions are the least aligned. is lower for the truncated directions in every case at every probe point of both pipelines, as predicted by Prop. 4.3. On Tiny-ImageNet they are also more often opposed on every seed throughout. On DDPM they are more often opposed before unlearning; during unlearning, opposition grows in every decile, most in the top one, so this sign test is mixed from step 250 on (end-point counts in the table).
- Truncation costs little progress. The kept directions carry 86–94% of the first-order progress on each loss; the truncated ones hold only 6–8% (Tiny-ImageNet) or 13–15% (DDPM) of .
- The leading directions carry the unlearning update. On Tiny-ImageNet, keeping as many directions from the bottom of the spectrum instead of the top raises the Avg. Gap from 5.35 to 26.49.
Implementation details of the analysis (Q3). We run the analysis on Tiny-ImageNet and DDPM, because in these two pipelines each SSU step filters one combined gradient, which is the setting analysed in Sec. 4. On Tiny-ImageNet, forget and retain images share each mini-batch, and on DDPM, each step uses one joint loss. We measure the gradients before unlearning, at six points along the unlearning run, and on the final checkpoints of the reported runs. All measurements use fixed probe batches: four mixed batches of 256 images per seed on Tiny-ImageNet, and eight pairs of 128 forget and 128 retained images per class on DDPM, with fixed noise and timesteps. For every filtered matrix, we compute , and with three backward passes through the same graph, using the losses and weights of the unlearning step. Since is then the gradient that SSU filters at that step, the kept directions are exactly those SSU keeps.
We thank the reviewer for their valuable feedback and will incorporate these clarifications into the revision.
Response to Reviewer 6tAr (1/3)
We sincerely thank Reviewer 6tAr for their feedback. We answer W1 (with Q1, Q2 and the minor point) in our first two comments and W2 (with Q3) in the third comment, and sincerely hope the reviewer will consider raising the score if our reply help address some of their concerns.
1. Definitions of Lf and Lr (W1, Q1). We respectfully clarify that in every SSU experiment both terms are minimized, with no sign flip; they differ in data and targets and, for the diffusion models and LLMs, also in form. The objective is Lu=cfLf+crLr, where the weights cf and cr follow existing works: SalUn [1] for classification, DDPM and Stable Diffusion, and Yuan et al. [2] for TOFU. The two terms are defined as follows (sg: stop-gradient; xt: noised input at step t):
Both losses are explicit, so Gf and Gr can be computed separately per weight matrix at any iterate, as in our response 5. We will clarify these definitions, including that both terms are minimized in every task, in the revised version.
2. The index ℓ, thresholds and number of layers (W1). ℓ indexes the weight matrices (layers). G(ℓ)=∇W(ℓ)Lu is reshaped to mℓ×nℓ (a convolution kernel to out × in·kHkW) and keeps kℓ=max(1,⌊γmin(mℓ,nℓ)⌋) components. We respectfully clarify that γ is a single fixed number per run, shared by all layers: nothing is tuned per layer, and kℓ varies only with the matrix size. The number of hyperparameters therefore does not grow with depth. We will define ℓ, kℓ and the shared γ in Algorithm 1.
[1] Fan et al., SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generation, ICLR 2024, arXiv:2310.12508.
[2] Yuan et al., A Closer Look at Machine Unlearning for Large Language Models, ICLR 2025, arXiv:2410.08109.