- Rating
- 3(+1)
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
The paper proposes TRIM, a structured pruning method for SNNs that requires no fine-tuning. It selects channels and compensates the remaining weights by matching the mean and linear trend of surrogate membrane potentials.
Critical strengths
The paper addresses a practical problem, compressing pretrained SNNs without expensive retraining. Considering channel selection together with weight compensation is sensible. The experiments cover both vision and language models, with clear improvements over the full-trajectory baseline in several VGG settings.
Critical weaknesses
1.Table 7 lists two time steps for ResNet20. In this case, the two temporal bases retain the full temporal information, so TRIM and the baseline should be equivalent under identical settings. Their different results in Table 1 need explanation. 2.The paper should specify whether methods use identical calibration samples and report repeated runs, especially where the improvements are small. 3.Changing the objective affects both channel selection and weight compensation. The current analysis does not separate these effects. 4.The comparison with the full-trajectory baseline is useful, but it does not sufficiently justify the choice of mean and linear trend. Comparisons with mean-only reconstruction, alternative two-dimensional bases, and higher-dimensional subspaces would clarify the contribution of each design choice.
Questions for authors
See Weakness
6 comments
Response to W2
W2: Do the methods use identical calibration samples, and are the reported improvements supported by repeated runs, especially when the differences are small?
We thank the reviewer for requesting this clarification. Calibration-set sizes were selected using the validation set, not the test set, and the same selection procedure was applied to TRIM and all competing methods. In the ResNet20 comparison described in W1, both methods used identical calibration samples and ordering within each seed. The additional ResNet19 comparison used the same calibration budget and seed set across methods.
We repeated pruning over 10 calibration seeds and report the available results below as top-1 accuracy (%) with mean ± standard deviation. These results cover VGG16 on three datasets, SEW-ResNet18 on ImageNet, and ResNet19 and ResNet20 on both CIFAR datasets. These runs vary calibration seeds, rather than retraining the dense models.
Table R3. Repeated-run results on VGG16 and SEW-ResNet18.
| Model | Dataset | Pruning ratio | OBC | Baseline | TRIM |
|---|---|---|---|---|---|
| VGG16 | CIFAR-10 | 10% | 92.47 ± 0.07 | 92.67 ± 0.06 | 93.13 ± 0.04 |
| 20% | 90.81 ± 0.08 | 90.96 ± 0.07 | 91.36 ± 0.06 | ||
| 30% | 88.60 ± 0.07 | 87.55 ± 0.06 | 88.91 ± 0.07 | ||
| 40% | 81.97 ± 0.06 | 81.13 ± 0.05 | 84.10 ± 0.06 | ||
| 50% | 69.41 ± 0.13 | 68.47 ± 0.09 | 74.42 ± 0.14 | ||
| VGG16 | CIFAR-100 | 10% | 69.07 ± 0.08 | 69.40 ± 0.07 | 69.61 ± 0.06 |
| 20% | 64.81 ± 0.09 | 64.30 ± 0.08 | 65.81 ± 0.08 | ||
| 30% | 58.46 ± 0.07 | 55.33 ± 0.08 | 58.54 ± 0.06 | ||
| 40% | 45.76 ± 0.08 | 39.08 ± 0.07 | 48.52 ± 0.07 | ||
| VGG16 | TinyImageNet | 10% | 53.91 ± 0.06 | 55.16 ± 0.04 | 55.57 ± 0.05 |
| 20% | 44.95 ± 0.05 | 47.52 ± 0.06 | 49.91 ± 0.07 | ||
| 30% | 30.61 ± 0.08 | 34.83 ± 0.07 | 41.42 ± 0.08 | ||
| SEW-ResNet18 | ImageNet | 10% | 59.97 ± 0.21 | 59.77 ± 0.16 | 59.92 ± 0.19 |
| 20% | 54.86 ± 0.13 | 54.68 ± 0.14 | 55.52 ± 0.15 | ||
| 30% | 47.92 ± 0.14 | 48.51 ± 0.15 | 49.82 ± 0.13 | ||
| 40% | 36.79 ± 0.18 | 40.47 ± 0.16 | 42.34 ± 0.17 | ||
| 50% | 23.61 ± 0.19 | 29.22 ± 0.15 | 31.87 ± 0.16 |
The larger improvements remain well above the observed seed-to-seed variation. On VGG16/CIFAR-10 at 10% pruning, TRIM exceeds OBC and the baseline by 0.66 and 0.46 percentage points, respectively. At 20% pruning on TinyImageNet, the corresponding gains are 4.96 and 2.39 points; at 50% pruning on ImageNet, they are 8.26 and 2.65 points. Thus, the improvements are not limited to small differences in individual runs.
We also distinguish these gains from close comparisons. On ImageNet at 10% pruning, OBC and TRIM achieve 59.97 ± 0.21% and 59.92 ± 0.19%, respectively. Similarly, TRIM's 0.08-point lead over OBC on VGG16/CIFAR-100 at 30% pruning is small. We do not interpret these cases as clear advantages for TRIM.
Table R4. Additional ResNet19 results, T=4.
| Pruning ratio | CIFAR-10 | CIFAR-100 | ||||
|---|---|---|---|---|---|---|
| OBC | Baseline | TRIM | OBC | Baseline | TRIM | |
| 10% | 94.27 ± 0.08 | 94.34 ± 0.07 | 94.41 ± 0.07 | 74.28 ± 0.10 | 74.40 ± 0.11 | 74.53 ± 0.09 |
| 20% | 93.55 ± 0.06 | 93.75 ± 0.07 | 94.05 ± 0.06 | 73.22 ± 0.15 | 73.53 ± 0.15 | 73.75 ± 0.15 |
| 30% | 92.71 ± 0.07 | 93.07 ± 0.06 | 93.45 ± 0.07 | 72.01 ± 0.15 | 72.34 ± 0.13 | 72.70 ± 0.15 |
| 40% | 91.67 ± 0.07 | 92.01 ± 0.07 | 92.71 ± 0.07 | 69.15 ± 0.16 | 70.16 ± 0.15 | 71.09 ± 0.17 |
| 50% | 88.89 ± 0.07 | 89.73 ± 0.09 | 90.68 ± 0.08 | 66.76 ± 0.15 | 67.63 ± 0.14 | 68.92 ± 0.13 |
TRIM has higher mean accuracy at every evaluated ratio on both datasets. At 50% pruning, its gains over OBC are 1.79 and 2.16 points on CIFAR-10 and CIFAR-100, respectively, and its gains over the baseline are 0.95 and 1.29 points. These repeated experiments extend the evidence to an additional residual architecture.
Table R5. ResNet20 results, T=2.
| Pruning ratio | CIFAR-10 | CIFAR-100 | ||
|---|---|---|---|---|
| Baseline | TRIM | Baseline | TRIM | |
| 10% | 95.90 ± 0.12 | 95.92 ± 0.13 | 77.98 ± 0.16 | 78.01 ± 0.17 |
| 20% | 95.57 ± 0.11 | 95.56 ± 0.12 | 76.28 ± 0.15 | 76.25 ± 0.16 |
| 30% | 95.12 ± 0.13 | 95.14 ± 0.11 | 73.85 ± 0.18 | 73.90 ± 0.16 |
| 40% | 93.92 ± 0.13 | 93.89 ± 0.14 | 69.98 ± 0.16 | 69.95 ± 0.16 |
| 50% | 91.87 ± 0.12 | 91.85 ± 0.13 | 63.30 ± 0.16 | 63.90 ± 0.17 |
As discussed in W1, the T=2 objectives are mathematically equivalent, so these accuracy differences are not evidence of benefits from temporal-subspace selection. We include the repeated results for completeness and keep this implementation-consistency check separate from the T>2 comparisons.
We will include these repeated-run results in the revised manuscript and clarify the calibration-set selection and sampling protocols. Reporting means and standard deviations will make the substantial gains and close comparisons explicit.
Response to W3
W3: Changing the reconstruction objective affects both channel selection and weight compensation. The current analysis does not separate their contributions.
We thank the reviewer for suggesting this analysis. We evaluated all four combinations of channel selection and weight compensation using the full-trajectory baseline and TRIM. This allows us to compare compensation objectives with the selected channels fixed, and channel selections under the same compensation objective. The experiments cover VGG16 and ResNet19 on CIFAR-100, both with T=4.
For each cross-combination, we fixed the channel sets selected by the source method and recomputed compensation from the original pretrained checkpoint, without reselecting channels or fine-tuning.
Table R6. Channel-selection and weight-compensation ablation on CIFAR-100. Top-1 accuracy (%) at different pruning ratios.
| Model | Channel selection | Weight compensation | 10% | 20% | 30% | 40% | 50% |
|---|---|---|---|---|---|---|---|
| VGG16 | Baseline | Baseline | 69.41 | 64.26 | 55.17 | 39.03 | — |
| TRIM | Baseline | 69.47 | 64.35 | 55.48 | 41.26 | — | |
| Baseline | TRIM | 69.52 | 65.55 | 57.32 | 45.49 | — | |
| TRIM | TRIM | 69.57 | 65.78 | 58.50 | 48.53 | — | |
| ResNet19 | Baseline | Baseline | 74.40 | 73.53 | 72.34 | 70.16 | 67.63 |
| TRIM | Baseline | 74.33 | 73.57 | 72.36 | 70.28 | 67.97 | |
| Baseline | TRIM | 74.41 | 73.67 | 72.51 | 70.64 | 68.31 | |
| TRIM | TRIM | 74.53 | 73.75 | 72.70 | 71.09 | 68.92 |
The cross-combinations are additional ablation results; the complete baseline and TRIM entries are the previously reported references. “—” denotes a setting not reported here.
The compensation comparison shows that TRIM's gains do not depend solely on its channel choices. On VGG16 at 40% pruning, replacing baseline compensation with TRIM compensation improves accuracy from 39.03% to 45.49% with baseline-selected channels fixed, a gain of 6.46 percentage points. With TRIM-selected channels fixed, the same replacement improves accuracy from 41.26% to 48.53%, a gain of 7.27 points. Temporal-subspace compensation therefore improves the reported accuracy with either channel selection.
Channel selection also contributes. With TRIM compensation fixed, using TRIM-selected rather than baseline-selected channels improves VGG16 accuracy from 45.49% to 48.53%, a further 3.04 points. The ResNet19 comparison shows a similar pattern: at 50% pruning, replacing baseline compensation with TRIM compensation on baseline-selected channels improves accuracy from 67.63% to 68.31%; using TRIM-selected channels with the same compensation objective further improves it to 68.92%.
Across both architectures and all reported pruning ratios, baseline selection with TRIM compensation outperforms TRIM selection with baseline compensation, while complete TRIM achieves the highest reported accuracy. These results identify a substantial contribution from compensation and an additional benefit from channel selection, rather than attributing the improvement to either component alone. We will add this selection–compensation ablation to the experimental section of the revised manuscript.
Response to W1
W1: With T=2, the two temporal bases preserve all temporal information. Why do TRIM and the full-trajectory baseline produce different results in Table 1?
We thank the reviewer for identifying this special case. The reviewer is correct: at T=2, the normalized constant and linear components form a complete orthonormal basis. Projecting onto both components is therefore an orthogonal change of coordinates, not dimensionality reduction. With identical inputs, normalization, and regularization, TRIM and the full-trajectory baseline optimize equivalent objectives in exact arithmetic. The temporal-subspace distinction applies only when T>2.
The two implementations nevertheless follow different numerical paths. Both accumulate calibration statistics and apply compensating weight updates in FP32, but the baseline uses the full surrogate membrane-potential trajectories, whereas TRIM first projects them onto the temporal components. These algebraically equivalent operations need not produce identical floating-point results because their operation orders and rounding differ.
Our implementation checks found different FP32 statistics for identical inputs and different compensated weights even when early-layer channel selections agreed. These weight differences can change whether neurons near the firing threshold emit spikes, affecting the inputs used to calibrate subsequent layers. Meanwhile, small differences between closely ranked candidate scores can change the greedy deletion order. We observed subsequent differences in spikes and channel selections, consistent with numerical differences propagating through layer-wise pruning. Thus, identical precision and seeds do not guarantee identical pruned networks across the two implementations.
To assess the resulting variation, we repeated ResNet20 experiments on CIFAR-10 and CIFAR-100 with 10 calibration seeds and 4,096 samples. For each dataset, the pretrained checkpoint was fixed. Within each seed, both methods used identical calibration samples and ordering, with pruning and evaluation settings aligned. All results below are top-1 accuracy (%) reported as mean ± standard deviation over 10 seeds.
Table R1. ResNet20 at T=2.
TRIM does not achieve higher mean accuracy in every setting. More importantly, regardless of which implementation performs better, differences at T=2 cannot demonstrate benefits from temporal-subspace selection because no temporal component is discarded.
We therefore additionally evaluated ResNet19 at T=4, where the two retained components span a strictly lower-dimensional subspace. This comparison tests the method outside the equivalence case raised by the reviewer. For each dataset, OBC, the baseline, and TRIM used the same pretrained checkpoint, calibration budget, pruning ratios, and 10 calibration seeds.
Table R2. Additional ResNet19 results at T=4.
At T=4, TRIM achieves higher mean accuracy than both alternatives at every evaluated pruning ratio on both datasets. At 50% pruning, it improves over the baseline by 0.95 and 1.29 percentage points on CIFAR-10 and CIFAR-100, respectively, and over OBC by 1.79 and 2.16 points. These results support TRIM in a setting where its temporal-subspace objective genuinely differs from full-trajectory reconstruction.
We will explicitly state the T=2 equivalence and the T>2 dimensionality reduction in Section 3.3, and revise the Table 1 discussion so that T=2 differences are not presented as evidence of subspace benefits. We will also include the repeated ResNet20 results and the additional ResNet19 comparison, with numerical precision, statistics computation, calibration sampling, and evaluation settings documented in the appendix.