Pith. sign in

REVIEW 4 major objections 6 minor 44 references

Pruning Everything, Everywhere, All at Once

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A greedy similarity rule — keep whichever candidate, layer-pruned or filter-pruned, is most CKA-similar to its parent — lets pruning remove both structures at once, reaching 95.82% FLOPs reduction on ResNet110 with a 2.91-point accuracy…

desk verdict A plausible pruning alternation idea whose decision rule is under-specified: the CKA comparison between parent and candidate is never defined across different depths and widths. read the letter →

arxiv 2506.04513 v1 pith:VTWJX2CN submitted 2025-06-04 cs.CV

Gustavo Henrique do Nascimento , Ian Pons , Anna Helena Reali Costa , Artur Jordao This is my paper
classification cs.CV
keywords structuredpruninglayerfilterCenteredKernelAlignmentrepresentationsimilarityFLOPsreductionResNetnetworkcompression
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to settle a decision that most pruning methods dodge: when compressing a network, should the next cut remove a layer or a filter? Its answer is to generate one candidate of each type, compare each candidate's internal representations with those of the unpruned parent using Centered Kernel Alignment (CKA), and keep the more similar one. Repeating this greedy choice, with brief fine-tuning between rounds, produces networks that are simultaneously shallower and narrower. If the claim holds, the payoff is compression beyond what layer-only or filter-only pruning reaches: 72.67% FLOPs reduction on ResNet56 with accuracy up 0.19 points, 86.37% with a 1.36-point drop, and 95.82% on ResNet110 with a 2.91-point drop, along with improved robustness to adversarial and out-of-distribution inputs. The authors position the rule as a structure-agnostic route to pruning that extends to ResNet50 on ImageNet and to Transformers on tabular data.

What carries the argument

Centered Kernel Alignment (CKA), the normalized Hilbert-Schmidt Independence Criterion between feature-map kernels of two networks, is the object that carries the decision: $$CKA(R, R_{F'}) = \frac{HSIC(R,R_{F'})}{\sqrt{HSIC(R,R) \cdot HSIC(R_{F'},R_{F'})}}.$$ It is used as the rule in Eq. (3): keep the layer-pruned candidate if $CKA(F,F'_l) \geq CKA(F,F'_f)$, otherwise keep the filter-pruned candidate. Both candidates are produced by the same KL-divergence pruning criterion, a short fine-tuning is applied before each comparison, and the selected candidate becomes the parent of the next iteration, giving an $O(\log n)$ search over the binary tree of possible pruning paths.

What would settle it

Fix a small architecture and data set, e.g., ResNet32 on CIFAR-10, and enumerate every layer-or-filter pruning path up to a target sparsity, fine-tuning each candidate. At each decision node, record which of the two candidates has higher post-fine-tuning accuracy and compare that with which candidate has higher CKA to its parent under an explicit layer- and channel-alignment rule. If the higher-CKA candidate is not the higher-accuracy candidate in a substantial fraction of nodes, the selection rule's predictive claim is false; because the paper does not specify the alignment, the test must state which alignment it used.

Watch

Extended reading notes

Core claim

The paper's central claim is that representation similarity to the parent is a sufficient guide for deciding which structure to prune next. At each iteration, two candidates are built from the current parent $F$ using the same pruning criterion: one by removing a layer, $F'_l$, and one by removing a filter, $F'_f$. After brief fine-tuning, the method computes $CKA(F,F'_l)$ and $CKA(F,F'_f)$ and keeps the layer candidate when the scores are equal, because layer removal also improves inference efficiency; otherwise it keeps the more similar candidate. The winner becomes the parent for the next round, so the process walks a binary decision tree of structural choices in $O(\log n)$ time. On the paper's evidence this rule yields accuracy-preserving or accuracy-improving pruned models at FLOPs reductions that single-structure baselines cannot reach, including 86.37% on ResNet56 and 95.82% on ResNet110.

Load-bearing premise

The method assumes that the similarity score between a parent network and a pruned candidate — computed from internal activity maps that no longer line up one-to-one after layers or channels are removed — reliably measures how much accuracy the candidate keeps, even though the paper does not specify exactly which layers are compared or how activations are matched.

Editorial extensions

If this is right

  • If the claim holds, structured pruning no longer has to commit to one structure: the CKA rule removes layers and filters in alternation, pushing compression past the layer-only ceiling of roughly 75.5% FLOPs on CIFAR-10/ResNet56.
  • If the claim holds, the rule transfers across datasets and architectures: it raises accuracy on ResNet32/44 and on Transformers over tabular data, matches strong baselines on ImageNet/ResNet50, and reaches 86.37% and 95.82% FLOPs reduction on ResNet56 and ResNet110 with small accuracy cost.
  • If the claim holds, high-sparsity pruned models also improve robustness rather than merely preserving accuracy: pruned ResNets improve FGSM adversarial robustness and CIFAR-C/CIFAR-10.2 out-of-distribution performance at around 90% FLOPs reduction.
  • If the claim holds, the decision rule is the contribution rather than any single score: the paper reports positive results with CKA, Linear, Gaussian Stochastic, and Wasserstein similarities and with both KL-divergence and ℓ1 pruning criteria.
  • If the claim holds, the FLOPs savings turn into estimated environmental and financial savings: up to 83.31% lower CO2 emissions and up to 83.67% lower estimated financial costs on the pruned ResNet models.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: because CKA needs only feature maps, the same selection rule could serve as a label-free reward signal for deciding where to prune or when to stop, a use the paper does not pursue.
  • Editorial inference: the paper reports FLOPs and estimated carbon, not measured latency or energy on specific hardware; a direct extension would test whether the CKA-chosen paths also minimize real runtime, which FLOPs do not guarantee.
  • Editorial inference: the unspecified layer/channel alignment between parent and candidate is the main extension point; once an alignment rule is fixed, the binary layer-versus-filter tree naturally widens into a multi-branch tree over blocks, heads, or residual branches, which the paper names only as future work.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes an iterative structured-pruning method that, at each step, generates two candidate subnetworks from a parent network—one by pruning layers and one by pruning filters—and selects the candidate whose internal representation is most similar to the parent's, as measured by Centered Kernel Alignment (CKA). The selected candidate becomes the parent for the next iteration, after a short fine-tuning step. The authors claim that this greedy, CKA-guided alternation between layer and filter pruning yields highly sparse models that preserve predictive accuracy, and they report state-of-the-art FLOP reductions on CIFAR-10 (e.g., 86.37% on ResNet56 and 95.82% on ResNet110), competitive results on ImageNet with ResNet50, and additional experiments on Transformers, robustness to adversarial/out-of-distribution samples, and carbon-emission reduction.

Significance. If the central claim holds, the paper addresses a genuine gap in structured pruning: most existing methods remove only layers or only filters, whereas this method systematically alternates between both. The reported FLOP reductions, especially 95.82% on ResNet110 with a limited accuracy drop, would be notable. The Random Walk comparison in Section IV provides a meaningful external control, and the public release of code is a strength. However, the missing specification of how CKA is computed across architectures of different depth and width, together with the absence of error bars and the loose protocol for choosing operating points, currently prevents full verification of the method and its claims.

major comments (4)
  1. [Section III, Eq. (2) and Algorithm 1] The selection rule in Eq. (3) compares CKA(F, F'_l) and CKA(F, F'_f), but the paper never states which feature maps are used when parent and candidate differ in depth (layer pruning removes a layer, changing layer indices) and width (filter pruning changes channel counts). Without an explicit alignment rule—for example, matching residual blocks by position, using only the final logits, or some interpolation—Eq. (3) is undefined, and the central claim that CKA-guided selection preserves predictive ability is not testable or reproducible. The authors must specify the alignment precisely and justify that it is not arbitrarily favoring one candidate type.
  2. [Section IV, Tables I and III] All main accuracy and FLOP results are reported as point estimates with no error bars, seeds, or variance, while the Random Walk baseline is averaged over only three runs. Given that the iterative process involves fine-tuning and stochasticity, the claimed improvements of +0.19 pp, +0.09 pp, and +0.66 pp over Random Walk in Table III could easily fall within run-to-run noise. The authors should report means and standard deviations over at least three to five independent seeds for both the proposed method and the Random Walk baseline.
  3. [Section IV, 'Is CKA Better than Random Choice?' and Table III] The Random Walk control shows that the choice of pruning path matters, but it does not establish that the CKA similarity signal is the operative cause of the improvement. A stronger and more direct baseline is to select the candidate with higher validation accuracy after the same fine-tuning procedure, or to compare against a criterion that directly estimates accuracy. Without such a baseline, the paper's claim that representation similarity is an effective proxy for predictive preservation remains indirect, since the accuracy gain might stem from other properties of the greedy path selection.
  4. [Section IV and Abstract] There are internal numerical inconsistencies that must be corrected: the text in Section IV says 'achieving a reduction of 86.57%' while the abstract and Table I report 86.37% for ResNet56; moreover, the 86.37% row has a delta of -1.36 pp, which is described in the abstract as a 'minimal drop' and elsewhere as 'negligible'. The authors should reconcile these numbers and characterize the accuracy drops more carefully, especially since -1.36 pp and -2.91 pp (ResNet110 at 95.82%) are not obviously negligible.
minor comments (6)
  1. [Algorithm 1 and Section IV] The 10-epoch pre-fine-tuning mentioned in Section IV is not shown in Algorithm 1; please include it or explicitly annotate the algorithm with a note that this step occurs before line 4.
  2. [Section III, Eq. (1)] The definition of HSIC uses m examples, but the feature maps from the parent and pruned networks may have different spatial dimensions; please clarify whether CKA is computed on aligned, per-example vectors of fixed size and how that alignment is obtained.
  3. [Throughout] There are typos and reference formatting issues, including 'Centred' vs 'Centered', 'Manson el al.' instead of 'Manson et al.', and several incomplete author lists such as 'A. D. et al.' and 'S. G. et al.'; these should be fixed in a final version.
  4. [Figure 1 caption] The caption says the ideal point is 'intentionally misused' and placed 'on the top right'; please clarify whether this is a mistake or a deliberate illustrative choice, and correct the notation so the axes are unambiguous.
  5. [Appendix VI-C] The Transformer experiments reportedly use 10 layers with '128 heads' each; for a tabular dataset this seems implausibly large and likely a typo. Please verify and correct the architecture description.
  6. [Section IV, 'The Role of Fine-tuning'] The sentence 'we discount the amount of 10-epoch from the remaining fine-tuning epochs' is unclear; please specify how the discounting is applied to the total fine-tuning budget.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the CKA-based selection rule is evaluated against an external Random Walk control, and no equation reduces the reported accuracy results to the CKA values themselves.

full rationale

The paper's central claim is that, among a layer-pruned and a filter-pruned candidate, the one with higher CKA similarity to the parent better preserves predictive ability. This is an empirical hypothesis, not a definitional identity: CKA is computed from feature maps (Eq. 2) while success is measured by task accuracy (Tables I-III), and the Random Walk baseline in Table III varies only the selection mechanism while holding candidate generation and fine-tuning fixed. No parameter is fitted so that CKA predicts accuracy, and no equation in the paper equates the reported delta-accuracy values with the CKA comparisons in Eq. 3. The self-citations to Pons et al. [6] and Jordao et al. [22] are used for motivation, iterative-pruning convention, and baseline comparison, not as an external uniqueness theorem or as the proof that CKA selection works. The main reproducibility concern raised by the skeptic—that the paper never specifies which feature maps are aligned when parent and candidate differ in depth or width, leaving Eq. 3 under-defined—is a correctness/definitional gap, not a circular reduction. I therefore find no significant circularity; the score reflects only the presence of minor self-citations that do not carry the argument.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The method is empirical and makes no new theoretical claims. It relies on a proxy metric (CKA), a greedy search over two structure types, an underspecified pruning schedule, and no new physical or architectural entities.

free parameters (6)
  • Per-iteration pruning magnitude = Not stated
    The number of layers or filters removed at each iteration is never quantified; it is a hand-chosen schedule that determines FLOPs reduction and accuracy.
  • Number of iterations K = Not stated
    Results are reported at selected K values, with no criterion given for stopping or for matching a target FLOPs budget.
  • Tie-breaking epsilon for layer preference = Not stated
    Equation 3 favors layer pruning on ties via an additive epsilon, but its value is not reported.
  • CKA feature layer and alignment = Not stated
    Which feature maps enter Eq. 2 and how layers are aligned between parent and pruned candidates are unspecified, making the similarity computation a free choice.
  • 10-epoch pre-fine-tuning = 10 epochs
    Adopted from prior work [8], [25]; the effect of this schedule on the CKA comparison is not ablated.
  • Capacity matching constraint = Not defined
    Candidates are said to have similar capacity following [22], but no operational definition is given; this controls the candidate generation step.
assumptions (5)
  • domain assumption CKA similarity between a candidate subnetwork and its parent is a valid proxy for preserving predictive ability.
    The entire selection rule in Eq. 3 is built on this; validation accuracy is not used for the decision.
  • domain assumption Greedy one-step selection by highest CKA leads to good global pruning trajectories.
    The method never compares the selected path against exhaustive search or lookahead; the authors state greedy choice is used for efficiency.
  • domain assumption The KL-divergence criterion scores filters and layers comparably.
    The same criterion c is applied to both structures, and the paper asserts it avoids pitfalls of filter-specific criteria such as L1 norm.
  • domain assumption Feature maps of a pruned network and its parent admit a meaningful CKA comparison.
    CKA requires aligned examples and kernels, but layer and channel mismatch after pruning is not resolved in the text.
  • domain assumption Fine-tuning for 10 epochs after each pruning step restores candidates to a comparable state.
    The 10-epoch schedule from prior work is adopted to make comparisons fair; no ablation of the schedule is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Pruning Everything, Everywhere, All at Once." pith.science (2026). https://pith.science/paper/VTWJX2CN

@misc{pith2026250604513,
  author       = {Pith},
  title        = {Pith review of: Pruning Everything, Everywhere, All at Once},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VTWJX2CN}},
  note         = {Machine review of arXiv:2506.04513}
}
read the original abstract

Deep learning stands as the modern paradigm for solving cognitive tasks. However, as the problem complexity increases, models grow deeper and computationally prohibitive, hindering advancements in real-world and resource-constrained applications. Extensive studies reveal that pruning structures in these models efficiently reduces model complexity and improves computational efficiency. Successful strategies in this sphere include removing neurons (i.e., filters, heads) or layers, but not both together. Therefore, simultaneously pruning different structures remains an open problem. To fill this gap and leverage the benefits of eliminating neurons and layers at once, we propose a new method capable of pruning different structures within a model as follows. Given two candidate subnetworks (pruned models), one from layer pruning and the other from neuron pruning, our method decides which to choose by selecting the one with the highest representation similarity to its parent (the network that generates the subnetworks) using the Centered Kernel Alignment metric. Iteratively repeating this process provides highly sparse models that preserve the original predictive ability. Throughout extensive experiments on standard architectures and benchmarks, we confirm the effectiveness of our approach and show that it outperforms state-of-the-art layer and filter pruning techniques. At high levels of Floating Point Operations reduction, most state-of-the-art methods degrade accuracy, whereas our approach either improves it or experiences only a minimal drop. Notably, on the popular ResNet56 and ResNet110, we achieve a milestone of 86.37% and 95.82% FLOPs reduction. Besides, our pruned models obtain robustness to adversarial and out-of-distribution samples and take an important step towards GreenAI, reducing carbon emissions by up to 83.31%. Overall, we believe our work opens a new chapter in pruning.

Figures

Figures reproduced from arXiv: 2506.04513 by the authors.

Figure 1
Figure 1. Left. A binary decision tree of potential pruning outcomes and the paths (red arrows) to follow using our approach. The process starts with an unpruned deep network (parent), for which we generate two subnetworks (pruned): one from layer pruning and the other from filter pruning. Then, with a similarity metric (i.e., Centred Kernel Alignment – CKA), we evaluate the resulting subnetworks to determine and select the o… view at source ↗
Figure 2
Figure 2. Performance of our method pruning Transformer architectures. [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

44 extracted references · 43 canonical work pages

  1. [1]

    Meta-learning adaptable foundation models,

    “Meta-learning adaptable foundation models,” in ICLR, 2025. [Online]. Available: https://openreview.net/forum?id=h0pACOIFxC

    work page 2025

  2. [2]

    The llama 3 herd of models,

    A. D. et al., “The llama 3 herd of models,” ArXiv, 2024

    work page 2024

  3. [3]

    LLMCarbon: Modeling the end-to-end carbon footprint of large language models,

    A. Faiz, S. Kaneda, R. Wang, R. C. Osi, P. Sharma, F. Chen, and L. Jiang, “LLMCarbon: Modeling the end-to-end carbon footprint of large language models,” in ICLR, 2024

    work page 2024

  4. [4]

    A survey on deep neural net- work pruning: Taxonomy, comparison, analysis, and recommendations,

    H. Cheng, M. Zhang, and J. Q. Shi, “A survey on deep neural net- work pruning: Taxonomy, comparison, analysis, and recommendations,” TPAMI, 2024

    work page 2024

  5. [5]

    Structured pruning for deep convolutional neural networks: A survey,

    Y . He and L. Xiao, “Structured pruning for deep convolutional neural networks: A survey,” TPAMI, 2023

    work page 2023

  6. [6]

    Effective layer pruning through similarity metric perspective,

    I. Pons, B. Yamamoto, A. H. R. Costa, and A. Jordao, “Effective layer pruning through similarity metric perspective,” in ICPR, 2024

    work page 2024

  7. [7]

    Layermerge: Neural network depth compression through layer pruning and merging,

    J. Kim, M. E. Halabi, M. Ji, and H. O. Song, “Layermerge: Neural network depth compression through layer pruning and merging,” ICML, 2024

    work page 2024

  8. [8]

    What makes a good prune? maximal unstructured pruning for maximal cosine similarity,

    G. Mason-Williams and F. Dahlqvist, “What makes a good prune? maximal unstructured pruning for maximal cosine similarity,” in ICLR, 2024

    work page 2024

Show all 44 references
  1. [9]

    Similarity of neural network representations revisited,

    S. Kornblith, M. Norouzi, H. Lee, and G. E. Hinton, “Similarity of neural network representations revisited,” in ICML, 2019

    2019

  2. [10]

    Compact language models via pruning and knowledge distillation,

    S. Muralidharan, S. T. Sreenivas, R. B. Joshi, M. Chochowski, M. Pat- wary, M. Shoeybi, B. Catanzaro, and et al., “Compact language models via pruning and knowledge distillation,” in NeurIPS, 2024

    2024

  3. [11]

    Structural pruning via latency-saliency knapsack,

    M. Shen, H. Yin, P. Molchanov, L. Mao, J. Liu, and J. M. ´Alvarez, “Structural pruning via latency-saliency knapsack,” in NeurIPS, 2022

    2022

  4. [12]

    Jointly training and pruning cnns via learnable agent guidance and alignment,

    A. Ganjdanesh, S. Gao, and H. Huang, “Jointly training and pruning cnns via learnable agent guidance and alignment,” in CVPR, 2024

    2024

  5. [13]

    Quantifying the carbon emissions of machine learning,

    A. Lacoste, A. Luccioni, V . Schmidt, and T. Dandres, “Quantifying the carbon emissions of machine learning,” in NeurIPS, 2019

    2019

  6. [14]

    Holistically evaluating the environmental impact of creating language models,

    “Holistically evaluating the environmental impact of creating language models,” in ICLR, 2025. [Online]. Available: https://openreview.net/forum?id=04qx93Viwj

    2025

  7. [15]

    Bilevelpruning: Unified dynamic and static channel pruning for convolutional neural networks,

    S. G. et al., “Bilevelpruning: Unified dynamic and static channel pruning for convolutional neural networks,” in CVPR, 2024

    2024

  8. [16]

    Auto-train-once: Controller network guided automatic network pruning from scratch,

    X. Wu, S. Gao, Z. Zhang, Z. Li, R. Bao, Y . Zhang, X. Wang, and H. Huang, “Auto-train-once: Controller network guided automatic network pruning from scratch,” in CVPR, 2024

    2024

  9. [17]

    Laco: Large language model pruning via layer collapse,

    Y . Yang, Z. Cao, and H. Zhao, “Laco: Large language model pruning via layer collapse,” EMNLP, 2024

    2024

  10. [18]

    The unreasonable ineffectiveness of the deeper layers,

    “The unreasonable ineffectiveness of the deeper layers,” in ICLR, 2025. [Online]. Available: https://openreview.net/forum?id=ngmEcEer8a

    2025

  11. [19]

    Shortened LLaMA: A simple depth pruning for large language models,

    B.-K. K. et al., “Shortened LLaMA: A simple depth pruning for large language models,” in ICLRW, 2024

    2024

  12. [20]

    Revisiting random channel pruning for neural network compression,

    Y . L. et al., “Revisiting random channel pruning for neural network compression,” in CVPR, 2022

    2022

  13. [21]

    Measuring statistical dependence with hilbert-schmidt norms,

    A. Gretton, O. Bousquet, A. J. Smola, and B. Sch ¨olkopf, “Measuring statistical dependence with hilbert-schmidt norms,” in ALT, 2005

    2005

  14. [22]

    When layers play the lottery, all tickets win at initialization,

    A. Jordao, G. C. de Ara ´ujo, H. de Almeida Maia, and H. Pedrini, “When layers play the lottery, all tickets win at initialization,” in ICCV, 2023

    2023

  15. [23]

    Neural network pruning with residual-connections and limited-data,

    J.-H. Luo and J. Wu, “Neural network pruning with residual-connections and limited-data,” in CVPR, 2020

    2020

  16. [24]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in CVPR, 2016

    2016

  17. [25]

    A simple and effective pruning approach for large language models,

    M. Sun, Z. Liu, A. Bair, and J. Z. Kolter, “A simple and effective pruning approach for large language models,” ICLR, 2024

    2024

  18. [26]

    Shallowing deep networks: Layer-wise pruning based on feature representations,

    S. Chen and Q. Zhao, “Shallowing deep networks: Layer-wise pruning based on feature representations,” TPAMI, 2019

    2019

  19. [27]

    Evolutionary shallowing deep neural networks at block levels,

    Y . Zhou, G. G. Yen, and Z. Yi, “Evolutionary shallowing deep neural networks at block levels,” TNNLS, 2022

    2022

  20. [28]

    DECORE: deep compression with reinforcement learning,

    M. Alwani, Y . Wang, and V . Madhavan, “DECORE: deep compression with reinforcement learning,” in CVPR, 2022

    2022

  21. [29]

    SOKS: automatic searching of the optimal kernel shapes for stripe-wise network pruning,

    G. Liu, K. Zhang, and M. Lv, “SOKS: automatic searching of the optimal kernel shapes for stripe-wise network pruning,” TPAMI, 2023

    2023

  22. [30]

    Revisit kernel pruning with lottery regulated grouped convolutions,

    S. Zhong, G. Zhang, N. Huang, and S. Xu, “Revisit kernel pruning with lottery regulated grouped convolutions,” in ICLR, 2022

    2022

  23. [31]

    On the channel pruning using graph convolution network for convolutional neural network acceleration,

    D. J. et al., “On the channel pruning using graph convolution network for convolutional neural network acceleration,” in IJCAI, 2022

    2022

  24. [32]

    Pruning neural networks via coresets and convex geometry: Towards no assumptions,

    M. Tukan, L. Mualem, and A. Maalouf, “Pruning neural networks via coresets and convex geometry: Towards no assumptions,” 2022

    2022

  25. [33]

    Topology-aware network pruning using multi-stage graph embedding and reinforcement learning,

    S. Y . et al., “Topology-aware network pruning using multi-stage graph embedding and reinforcement learning,” in ICML, 2022

    2022

  26. [34]

    Carrying out CNN channel pruning in a white box,

    Y . Zhang, M. Lin, C. Lin, J. Chen, and et al., “Carrying out CNN channel pruning in a white box,” TNNLS, 2023

    2023

  27. [35]

    Pruning networks with cross-layer ranking & k-reciprocal nearest filters,

    M. L. et al., “Pruning networks with cross-layer ranking & k-reciprocal nearest filters,” TNNLS, 2023

    2023

  28. [36]

    DAIS: automatic channel pruning via differentiable annealing indicator search,

    Y . G. et al., “DAIS: automatic channel pruning via differentiable annealing indicator search,” TPAMI, 2023

    2023

  29. [37]

    SOSP: efficiently capturing global correlations by second- order structured pruning,

    M. N. et al., “SOSP: efficiently capturing global correlations by second- order structured pruning,” in ICLR, 2022

    2022

  30. [38]

    Generalized shape metrics on neural representations,

    A. H. Williams, E. Kunz, S. Kornblith, and S. W. derman, “Generalized shape metrics on neural representations,” in NeurIPS, 2021

    2021

  31. [39]

    Representational dissimilarity metric spaces for stochastic neural networks,

    L. R. Duong, J. Zhou, J. Nassar, J. Berman, J. Olieslagers, and A. H. Williams, “Representational dissimilarity metric spaces for stochastic neural networks,” ICLR, 2023

    2023

  32. [40]

    Adversarial attack robust dataset pruning,

    “Adversarial attack robust dataset pruning,” in ICLR, 2025. [Online]. Available: https://openreview.net/forum?id=mORwTTZfWq

    2025

  33. [41]

    Adaptive sharpness-aware pruning for robust sparse networks,

    A. Bair, H. Yin, M. Shen, P. Molchanov, and J. M. Alvarez, “Adaptive sharpness-aware pruning for robust sparse networks,” in ICLR, 2024

    2024

  34. [42]

    Benchmarking neural network robustness to common corruptions and perturbations,

    D. Hendrycks and T. G. Dietterich, “Benchmarking neural network robustness to common corruptions and perturbations,” in ICLR, 2019

    2019

  35. [43]

    Harder or different? a closer look at distribution shift in dataset reproduction,

    S. L. et al., “Harder or different? a closer look at distribution shift in dataset reproduction,” in ICML, 2020

    2020

  36. [44]

    Human activity recognition based on smartphone and wearable sensors using multiscale dcnn ensemble,

    J. Sena, J. Barreto, C. Caetano, G. Cramer, and W. R. Schwartz, “Human activity recognition based on smartphone and wearable sensors using multiscale dcnn ensemble,” Neurocomputing, 2021. VI. A PPENDIX A. Robustness to Pruning Criteria In this experiment, we highlight the pote...

    2021

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.