Pith. sign in

REVIEW 3 major objections 4 minor 48 references

Exploiting Channel Similarity for Accelerating Deep Convolutional Neural Networks

T0 review · 3 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Pith's one-line read The paper argues that convolutional channels with similar batch-normalization statistics are redundant, letting a hierarchical-clustering prune remove 45.9% of ResNet-50's FLOPs at a 0.34% Top-5 accuracy cost.

desk verdict A simple, usable BN-statistics channel pruning heuristic with competitive results, but the distance metric's independence assumption is load-bearing and never quantitatively validated on the ImageNet regime. read the letter →

arxiv 1908.02620 v1 pith:CH6QMPS4 submitted 2019-08-06 cs.LG cs.CVstat.ML

classification cs.LGcs.CVstat.ML
keywords channelpruningsimilaritybatchnormalizationhierarchicalclusteringnetworkaccelerationrepresentationalredundancyResNet-50ImageNet
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 establish that redundant channels in a deep CNN are best identified by similarity of feature information rather than by magnitude of weights or activations. It derives a cheap similarity metric from batch-normalization parameters, groups channels with hierarchical clustering, and removes all but one representative per cluster. The paper claims this yields a pruned ImageNet ResNet-50 that loses only 0.34% Top-5 accuracy while shedding 45.90% of FLOPs, and that at a 30% FLOPs cut the pruned model actually beats the baseline's Top-1 accuracy. The broader payoff is a pruning recipe that needs no sparsity training or data-driven optimization and works directly on pretrained networks.

What carries the argument

The central object is the batch-normalization-derived channel distance $\mathrm{Dist}(N^{(i)},N^{(j)}) = (\beta^{(i)}-\beta^{(j)})^2 + (\gamma^{(i)})^2 + (\gamma^{(j)})^2$, treating each channel's normalized activations as i.i.d. random variables with mean $\beta$ and variance $\gamma^2$. This distance matrix feeds an agglomerative hierarchical clustering with one global threshold per network; after normalizing each layer's matrix to $[0,1]$, the algorithm keeps one representative channel per cluster and discards the rest. The theoretical backbone is Proposition 2, which upper-bounds the shift induced on the next layer's activations by $\lambda$ times the distance to the nearest remaining channel, where $\lambda$ is a small coefficient depending on kernel size and norm. This replaces an $O(H W B C^2)$ data-dependent comparison with an $O(C^2)$ computation using only BN parameters.

What would settle it

Keep a channel's $\beta$ and $\gamma$ fixed but permute its spatial activation patterns, or exchange it with another channel having the same $(\beta,\gamma)$ but different edge-selectivity; if the next layer's response shifts substantially even though Equation 4 is unchanged, then the BN-only distance misses spatial structure and the redundancy claim fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that redundant channels in a convolutional network are not the small-magnitude ones but the similar ones: two channels whose batch-normalized activations have nearly the same distribution carry overlapping feature information. For batch-normalized channels, the expected squared distance between channels converges in probability to $(\beta^{(i)}-\beta^{(j)})^2+(\gamma^{(i)})^2+(\gamma^{(j)})^2$, so the learned shift and scale parameters $\beta,\gamma$ alone define a channel-similarity metric. The paper then claims that channels clustered by this metric are functionally redundant, and that deleting all but the cluster member with the largest $|\gamma|$, followed by a short fine-tune, does not meaningfully reduce representational power. It supports this with a bound showing that removing one channel and merging its kernel into a similar channel shifts the next layer's activations at most by a small layer-dependent factor times the original channel distance.

Load-bearing premise

The load-bearing assumption is that two channels whose batch-normalization mean and scale parameters are close are truly interchangeable, so spatial structure and downstream effects can be ignored when deciding what to prune.

Editorial extensions

If this is right

  • On ImageNet, a pretrained ResNet-50 pruned with threshold $t=0.10$ drops only 0.34% Top-5 accuracy while removing 45.90% of FLOPs.
  • At a gentler 30% FLOPs reduction, the pruned ResNet-50 gains 0.08% Top-1 accuracy over the uncompressed baseline.
  • Because the metric uses only BN parameters, the pruning is one-shot on a pretrained model and does not require training with sparsity constraints, unlike SSL or NS.
  • The probabilistic distance estimate is stable across data batches and batch sizes, whereas direct activation-value distances fluctuate with the input sample.
  • Normalizing each layer's distance matrix lets a single global threshold automatically set per-layer pruning ratios, avoiding hand-designed target architectures.

Reading between the lines

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

  • The paper leaves implicit that the same beta-gamma distance could serve as a cheap layer-wise width-allocation signal during architecture search, or as a live duplication detector during training.
  • Since the proof bounds only a single-layer, one-to-one substitution while the algorithm prunes many channels across all layers and then fine-tunes, the method's practical success likely depends on fine-tuning repairing multi-layer coupling; testing this directly would clarify where the gain comes from.
  • For BN-free architectures, the metric would need an estimated analogue from activation moments, so portability beyond batch-normalized CNNs is a natural next test.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes a channel-pruning method for pre-trained CNNs based on the idea that channels with similar feature information are functionally redundant. It derives a distance between two channels from probabilistic modeling, approximating the expected squared activation difference under i.i.d. assumptions; using batch-normalization parameters, the distance reduces to (β_i−β_j)^2 + γ_i^2 + γ_j^2 (Eq. 4). The algorithm then clusters channels per layer with hierarchical clustering, retains one representative channel per cluster (largest |γ|), and fine-tunes the resulting model. The paper reports strong empirical results, including ResNet-50 on ImageNet with 45.90% FLOPs reduction at only 0.34% Top-5 accuracy drop, and 30% FLOPs reduction with slightly improved Top-1 accuracy. It also provides a theoretical bound (Proposition 2) for a single-layer, one-channel substitution.

Significance. If the proposed metric is valid, the method is practically attractive: it requires no sparsity training, no complex data-driven optimization, can be applied directly to pre-trained models, and uses a single global threshold to control pruning across all layers. The empirical results on CIFAR and ImageNet are competitive with or better than several existing channel-pruning methods. The derivation of the distance is principled under the stated assumptions, and the paper makes a good-faith effort to validate the approximation against activation-based distances (Section 4.3, Figure 3). However, the load-bearing assumption that per-channel batch-normalization parameters fully characterize functional similarity is not quantitatively validated, and the theoretical support is narrower than the actual algorithm. These issues affect the central claim, so the manuscript needs revision before the results can be considered fully supported.

major comments (3)
  1. [Section 3.2, Eq. (4); Proposition 1 (Eq. 2)] The distance formula omits the cross-channel covariance term. Under the stated independence assumption, E[(N_i−N_j)^2] = (β_i−β_j)^2 + γ_i^2 + γ_j^2, but without independence the exact expansion contains −2Cov(N_i,N_j). Batch-normalization parameters β and γ are per-channel and cannot encode this covariance. For example, if β_i=β_j and γ_i=γ_j, Eq. 4 gives 2γ^2, whereas the actual expected squared difference is 2γ^2−2Cov; this can be much larger for negatively correlated channels and near zero for positively correlated near-duplicates. Since the hierarchical clustering and all pruning decisions are driven entirely by this distance, the central claim that most channels in a similarity group can be removed is not secured by the provided derivation. The validation in Figure 3 compares Eq. 4 with Eq. 1 only visually for VGG-16 on CIFAR-10, and does not quantify rank agreement or test the ResNet-50/ImageNet regime where the headline results are reported. Please provide a quantitative comparison (e.g., Spearman rank correlation between Eq. 1 and Eq. 4 distances) on ImageNet layers, and either relax the independence assumption or incorporate a covariance-corrected term.
  2. [Section 3.3, Proposition 2 (Eq. 8)] Proposition 2 bounds only a single-layer, one-to-one substitution of one channel by one similar channel using a specifically updated kernel. The actual algorithm in Section 3.4 removes all but one channel per cluster, in every layer, and relies on fine-tuning to restore accuracy; the proof does not bound the cumulative effect of many removals, cross-layer propagation, or the interaction with the channel-selection layers inserted in residual blocks (Appendix A). As stated, Proposition 2 supports the weaker statement that removing one channel is safe if a sufficiently similar channel exists, not the abstract's claim that 'most channels within each such similarity group can be removed.' Please either extend the analysis to multi-channel removal or explicitly present Proposition 2 as heuristic motivation rather than a proof of the algorithm's safety.
  3. [Section 4.3, Figure 3] The validation of the probabilistic distance is weakened by the instability shown in the left panel of Figure 3: activation-based distances vary substantially across random batches, so comparing a 20-trial average to the BN-based distance does not establish that Eq. 4 is a reliable proxy for the distance on a given input. The right panel shows only visual agreement and no numerical error or rank-correlation statistic. Since the metric is the load-bearing component, please report a quantitative measure (e.g., mean absolute error, Spearman rank correlation) across multiple layers and architectures, including ResNet on ImageNet.
minor comments (4)
  1. [Section 3.2, Eq. (4)] The approximation symbol '⋍' is unusual; please define it explicitly or use a standard notation such as '≈'.
  2. [Appendix B] The numbering of propositions in the appendix is inconsistent with the main text: Proposition 1 appears as Proposition 3 and Proposition 2 appears as Proposition 4. Please align the numbering.
  3. [Table 1(c)] The column header 'Time Pruned' is ambiguous; clarify whether it reports absolute wall-clock time per batch or relative time saving, and specify the unit.
  4. [Figure 3 caption] The caption refers to the 'accuracy of the latter' for the red dotted line, but the left panel title and text describe the 'probabilistic approach'; use consistent terminology in the caption.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the BN-distance metric follows from the stated probabilistic assumptions, and the accuracy/FLOPs results are measured after pruning rather than being defined by the metric.

full rationale

The derivation chain is self-contained. Equation 4 is obtained algebraically from Proposition 1 under the paper's explicitly stated i.i.d. within-channel and mutual-independence assumptions; it is not fitted to pruning accuracy or to the reported ImageNet results. The hierarchical clustering uses this fixed distance metric with a global threshold t, but t controls pruning ratio, and the reported accuracies are measured after fine-tuning, so the headline results are empirical outcomes rather than consequences of the definition. Proposition 2 provides an independent mathematical upper bound on the distance shift from one-to-one channel substitution; it does not presuppose the success of the final pruning experiments. The paper does not rely on a self-citation chain or a uniqueness theorem to force its choice of metric, and its references to prior work such as [22] and [31] are external sources that motivate rather than define the result. The skeptic's concern that omitted inter-channel covariance weakens the metric is a validity or robustness critique of the independence assumption, not evidence that any claimed prediction reduces to its own input by construction. No load-bearing step is self-definitional, fitted-and-renamed, or forced by citation.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The method rests on learning-derived BN parameters, one user-set threshold, and a fine-tuning schedule. No new physical or mathematical entities are introduced.

free parameters (2)
  • Threshold t = 0.25, 0.225, 0.20, 0.19, 0.10, 0.06, 0.08, 0.11, 0.13 depending on dataset and model
    Global distance threshold that determines pruning ratio per layer; chosen post hoc to reach target FLOPs reductions, not predicted a priori. Reported accuracy is for these selected thresholds.
  • Fine-tuning epochs = 60 on CIFAR; ImageNet settings kept the same
    Hyperparameter for fine-tuning after pruning; standard but affects final accuracy.
assumptions (3)
  • domain assumption Activations belonging to the same channel are i.i.d. and activations from different channels are mutually independent (Proposition 1).
    Underlies the WLLN derivation of Eq. 2 and the BN-based distance in Eq. 4. The paper validates empirically in Section 4.3 but does not prove it.
  • domain assumption Batch-normalized activations follow a distribution with mean beta(c) and variance (gamma(c))^2.
    Assumes BN outputs are i.i.d. with these moments; ignores residual mean and variance of normalized inputs, though BN training encourages them to be 0 and 1.
  • ad hoc to paper The single-layer shift bound (Prop. 2) extends to full-network pruning with fine-tuning.
    The algorithm prunes all layers at once and fine-tunes, whereas the bound analyzes a single layer with one-to-one kernel compensation. The paper assumes multi-layer greedy pruning plus fine-tuning preserves accuracy despite the bound being local.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Exploiting Channel Similarity for Accelerating Deep Convolutional Neural Networks." pith.science (2026). https://pith.science/paper/CH6QMPS4

@misc{pith2026190802620,
  author       = {Pith},
  title        = {Pith review of: Exploiting Channel Similarity for Accelerating Deep Convolutional Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CH6QMPS4}},
  note         = {Machine review of arXiv:1908.02620}
}
read the original abstract

To address the limitations of existing magnitude-based pruning algorithms in cases where model weights or activations are of large and similar magnitude, we propose a novel perspective to discover parameter redundancy among channels and accelerate deep CNNs via channel pruning. Precisely, we argue that channels revealing similar feature information have functional overlap and that most channels within each such similarity group can be removed without compromising model's representational power. After deriving an effective metric for evaluating channel similarity through probabilistic modeling, we introduce a pruning algorithm via hierarchical clustering of channels. In particular, the proposed algorithm does not rely on sparsity training techniques or complex data-driven optimization and can be directly applied to pre-trained models. Extensive experiments on benchmark datasets strongly demonstrate the superior acceleration performance of our approach over prior arts. On ImageNet, our pruned ResNet-50 with 30% FLOPs reduced outperforms the baseline model.

Figures

Figures reproduced from arXiv: 1908.02620 by the authors.

Figure 1
Figure 1. Magnitude-based approach vs similarity-based approach to channel pruning. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Accuracy-acceleration curve on CIFAR dataset. Best viewed in color. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Left: Performance comparison between activation-value-based approach and its probabilis [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Left: Distribution of values in both the unnormalized distance matrix and the normalized [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Architecture of pruned VGG-16 model with 80% reduction of FLOPs on CIFAR-10 dataset. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Visualization of the channel distance matrix computed using different approaches. The [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 44 canonical work pages

  1. [1]

    Imagenet classification with deep convolutional neural networks

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. “Imagenet classification with deep convolutional neural networks”. In: Advances in neural information processing systems. 2012, pp. 1097–1105

  2. [2]

    Very deep convolutional networks for large-scale image recognition

    Karen Simonyan and Andrew Zisserman. “Very deep convolutional networks for large-scale image recognition”. In: International Conference on Learning Representations. 2015, pp. 1–14

  3. [3]

    Deep residual learning for image recognition

    Kaiming He et al. “Deep residual learning for image recognition”. In:Proceedings of the IEEE conference on computer vision and pattern recognition. 2016, pp. 770–778

  4. [4]

    Predicting parameters in deep learning

    Misha Denil et al. “Predicting parameters in deep learning”. In:Advances in neural information processing systems. 2013, pp. 2148–2156

  5. [5]

    Do deep nets really need to be deep?

    Jimmy Ba and Rich Caruana. “Do deep nets really need to be deep?” In:Advances in neural information processing systems. 2014, pp. 2654–2662

  6. [6]

    Learning Efficient Convolutional Networks through Network Slimming

    Zhuang Liu et al. “Learning Efficient Convolutional Networks through Network Slimming”. In: International Conference on Computer Vision. 2017, pp. 2755–2763

  7. [7]

    Channel pruning for accelerating very deep neural networks

    Yihui He, Xiangyu Zhang, and Jian Sun. “Channel pruning for accelerating very deep neural networks”. In: Proceedings of the IEEE International Conference on Computer Vision. 2017, pp. 1389–1397

  8. [8]

    Speeding Up Convolutional Neural Networks with Low Rank Expansions

    Max Jaderberg, Andrea Vedaldi, and Andrew Zisserman. “Speeding Up Convolutional Neural Networks with Low Rank Expansions”. In: British Machine Vision Conference. 2014

Show all 48 references
  1. [9]

    Exploiting linear structure within convolutional networks for efficient evaluation

    Emily L Denton et al. “Exploiting linear structure within convolutional networks for efficient evaluation”. In: Advances in neural information processing systems. 2014, pp. 1269–1277

  2. [10]

    Binarized neural networks: Training deep neural networks with weights and activations constrained to + 1 or -1

    Matthieu Courbariaux et al. “Binarized neural networks: Training deep neural networks with weights and activations constrained to + 1 or -1”. In: arXiv preprint arXiv:1602.02830 (2016)

  3. [11]

    Xnor-net: Imagenet classification using binary convolutional neural networks

    Mohammad Rastegari et al. “Xnor-net: Imagenet classification using binary convolutional neural networks”. In: European Conference on Computer Vision. Springer. 2016, pp. 525–542

  4. [12]

    Distilling the Knowledge in a Neural Network

    Geoffrey E. Hinton, Oriol Vinyals, and Jeffrey Dean. “Distilling the Knowledge in a Neural Network”. In: NIPS Deep Learning Workshop. 2014

  5. [13]

    Fitnets: Hints for thin deep nets

    Adriana Romero et al. “Fitnets: Hints for thin deep nets”. In: International Conference on Learning Representations. 2014

  6. [14]

    Pruning Filter via Geometric Median for Deep Convolutional Neural Net- works Acceleration

    Yang He et al. “Pruning Filter via Geometric Median for Deep Convolutional Neural Net- works Acceleration”. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2019

  7. [15]

    Optimal brain damage

    Yann LeCun, John S Denker, and Sara A Solla. “Optimal brain damage”. In: Advances in neural information processing systems. 1990, pp. 598–605

  8. [16]

    Second order derivatives for network pruning: Optimal brain surgeon

    Babak Hassibi and David G Stork. “Second order derivatives for network pruning: Optimal brain surgeon”. In: Advances in neural information processing systems. 1993, pp. 164–171

  9. [17]

    Learning both weights and connections for efficient neural network

    Song Han et al. “Learning both weights and connections for efficient neural network”. In: Advances in neural information processing systems. 2015, pp. 1135–1143

  10. [18]

    Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding

    Song Han, Huizi Mao, and William J Dally. “Deep compression: Compressing deep neural networks with pruning, trained quantization and huffman coding”. In: 2015

  11. [19]

    Dynamic network surgery for efficient dnns

    Yiwen Guo, Anbang Yao, and Yurong Chen. “Dynamic network surgery for efficient dnns”. In: Advances In Neural Information Processing Systems. 2016, pp. 1379–1387

  12. [20]

    Compressing Neural Networks using the Variational Information Bottleneck

    Bin Dai et al. “Compressing Neural Networks using the Variational Information Bottleneck”. In: Proceedings of the 35th International Conference on Machine Learning. 2018, pp. 1143– 1152. 9

  13. [21]

    Variational Dropout Sparsifies Deep Neural Networks

    Dmitry Molchanov, Arsenii Ashukha, and Dmitry P. Vetrov. “Variational Dropout Sparsifies Deep Neural Networks”. In: Proceedings of the 34th International Conference on Machine Learning. 2017, pp. 2498–2507

  14. [22]

    Data-free Parameter Pruning for Deep Neural Net- works

    Suraj Srinivas and R. Venkatesh Babu. “Data-free Parameter Pruning for Deep Neural Net- works”. In: Proceedings of the British Machine Vision Conference. 2015, pp. 31.1–31.12

  15. [23]

    Diversity networks: neural network compression using determi- nantal point processes

    Zelda Mariet and Suvrit Sra. “Diversity networks: neural network compression using determi- nantal point processes”. In: 2015

  16. [24]

    Clustering convolutional kernels to compress deep neural networks

    Sanghyun Son, Seungjun Nah, and Kyoung Mu Lee. “Clustering convolutional kernels to compress deep neural networks”. In: Proceedings of the European Conference on Computer Vision (ECCV). 2018, pp. 216–232

  17. [25]

    Learning structured sparsity in deep neural networks

    Wei Wen et al. “Learning structured sparsity in deep neural networks”. In:Advances in Neural Information Processing Systems. 2016, pp. 2074–2082

  18. [26]

    Fast convnets using group-wise brain damage

    Vadim Lebedev and Victor Lempitsky. “Fast convnets using group-wise brain damage”. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition . 2016, pp. 2554–2564

  19. [27]

    Data-driven sparse structure selection for deep neural networks

    Zehao Huang and Naiyan Wang. “Data-driven sparse structure selection for deep neural networks”. In: Proceedings of the European Conference on Computer Vision (ECCV). 2018, pp. 304–320

  20. [28]

    Less is more: Towards compact cnns

    Hao Zhou, Jose M Alvarez, and Fatih Porikli. “Less is more: Towards compact cnns”. In: European Conference on Computer Vision. Springer. 2016, pp. 662–677

  21. [29]

    Thinet: A filter level pruning method for deep neu- ral network compression

    Jian-Hao Luo, Jianxin Wu, and Weiyao Lin. “Thinet: A filter level pruning method for deep neu- ral network compression”. In: Proceedings of the IEEE international conference on computer vision. 2017, pp. 5058–5066

  22. [30]

    Discrimination-aware channel pruning for deep neural networks

    Zhuangwei Zhuang et al. “Discrimination-aware channel pruning for deep neural networks”. In: Advances in Neural Information Processing Systems. 2018, pp. 875–886

  23. [31]

    Reducing duplicate filters in deep neural networks

    Aruni RoyChowdhury et al. “Reducing duplicate filters in deep neural networks”. In: NIPS workshop on Deep Learning: Bridging Theory and Practice. V ol. 1. 2017

  24. [32]

    Shufflenet: An extremely efficient convolutional neural network for mobile devices

    Xiangyu Zhang et al. “Shufflenet: An extremely efficient convolutional neural network for mobile devices”. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018, pp. 6848–6856

  25. [33]

    Mobilenets: Efficient convolutional neural networks for mobile vision applications

    Andrew G Howard et al. “Mobilenets: Efficient convolutional neural networks for mobile vision applications”. In: arXiv preprint arXiv:1704.04861 (2017)

  26. [34]

    Designing Neural Network Architectures using Reinforcement Learning

    Bowen Baker et al. “Designing Neural Network Architectures using Reinforcement Learning”. In: 5th International Conference on Learning Representations. 2017

  27. [35]

    Neural Architecture Search with Reinforcement Learning

    Barret Zoph and Quoc V . Le. “Neural Architecture Search with Reinforcement Learning”. In: 5th International Conference on Learning Representations. 2017

  28. [36]

    Large-scale evolution of image classifiers

    Esteban Real et al. “Large-scale evolution of image classifiers”. In:Proceedings of the 34th International Conference on Machine Learning. 2017, pp. 2902–2911

  29. [37]

    Hierarchical Representations for Efficient Architecture Search

    Hanxiao Liu et al. “Hierarchical Representations for Efficient Architecture Search”. In: 6th International Conference on Learning Representations. 2018

  30. [38]

    Neural architecture optimization

    Renqian Luo et al. “Neural architecture optimization”. In: Advances in neural information processing systems. 2018, pp. 7816–7827

  31. [39]

    Darts: Differentiable architecture search

    Hanxiao Liu, Karen Simonyan, and Yiming Yang. “Darts: Differentiable architecture search”. In: arXiv preprint arXiv:1806.09055 (2018)

  32. [40]

    Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

    Sergey Ioffe and Christian Szegedy. “Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift”. In: Proceedings of the 32nd International Conference on Machine Learning. 2015, pp. 448–456

  33. [41]

    Deep sparse rectifier neural networks

    Xavier Glorot, Antoine Bordes, and Yoshua Bengio. “Deep sparse rectifier neural networks”. In: Proceedings of the 14th international conference on artificial intelligence and statistics. 2011, pp. 315–323

  34. [42]

    Densely connected convolutional networks

    Gao Huang et al. “Densely connected convolutional networks”. In:Proceedings of the IEEE conference on computer vision and pattern recognition. 2017, pp. 4700–4708

  35. [43]

    Rethinking the value of network pruning

    Zhuang Liu et al. “Rethinking the value of network pruning”. In: International Conference on Learning Representations. 2018. 10

  36. [44]

    Soft Filter Pruning for Accelerating Deep Convolutional Neural Networks

    Yang He et al. “Soft Filter Pruning for Accelerating Deep Convolutional Neural Networks”. In: Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence. 2018, pp. 2234–2240

  37. [45]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images. Tech. rep. Citeseer, 2009

  38. [46]

    Imagenet: A large-scale hierarchical image database

    Jia Deng et al. “Imagenet: A large-scale hierarchical image database”. In:Proceedings of the IEEE conference on computer vision and pattern recognition. 2009, pp. 248–255

  39. [47]

    Automatic differentiation in PyTorch

    Adam Paszke et al. “Automatic differentiation in PyTorch”. In: NIPS Workshop. 2017

  40. [48]

    Identity mappings in deep residual networks

    Kaiming He et al. “Identity mappings in deep residual networks”. In: European conference on computer vision. Springer. 2016, pp. 630–645. A. Implementation Implementation details. On CIFAR dataset, we make use of a variant of VGG-16 [6] and a 3-stage pre-activation ResNet [48]...

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.