Pith. sign in

REVIEW 4 major objections 4 minor 41 references

Group Pruning using a Bounded-Lp norm for Group Gating and Regularization

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

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

Pith's one-line read Exponential gates and a bounded-Lp penalty can drive unneeded channels exactly to zero during training, yielding high pruning rates with accuracy close to baseline.

desk verdict Useful pruning paper with competitive results, but the 'exact zero' claim for L2-regularized exponential gates is overstated and needs a precision caveat. read the letter →

arxiv 1908.03463 v1 pith:P52PUBAZ submitted 2019-08-09 stat.ML cs.LG

Chaithanya Kumar Mummadi , Tim Genewein , Dan Zhang , Thomas Brox , Volker Fischer This is my paper · ORCID
classification stat.MLcs.LG
keywords grouppruningnetworksparsityregularizationboundedLpnormexponentialgatingchannelmodelcompressiondeeplearning
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

This paper introduces two trainable mechanisms for channel-level pruning of deep neural networks: an exponential gating layer placed after convolutions, with trainable parameters that multiplicatively scale each channel, and a bounded variant of the L1 regularizer that saturates for large gate values. The central claim is that combining these tools makes insignificant channels become exactly zero during training, so pruning requires no threshold search and little to no fine-tuning. The authors demonstrate this on several architectures and datasets, reporting parameter reductions of 30% on ResNet-164, 69% on DenseNet-40, and 75% on MobileNetV2 on CIFAR100, with accuracy near baseline, plus competitive pruning of ResNet-50 and MobileNetV2 on ImageNet.

What carries the argument

The bounded-$\ell_{p,0}$ norm, defined coefficient-wise as $1 - \exp(-|x|^p/\sigma^p)$, interpolates between the $\ell_p$ and $\ell_0$ norms as $\sigma$ shrinks. Used as a regularizer on gate parameters with $p=1$ (bounded-$\ell_1$), it penalizes small gates strongly and saturates for large gate values. The exponential gating layer uses the same function with $p=2$ and $\sigma=1$ to produce gate values in $[0,1)$ that multiply each channel; because the gate activation and the regularizer both vanish only at a gate parameter of zero, channels can be driven exactly to zero. The paper also schedules $\sigma$ during training, gradually reducing it to strengthen regularization and steer insignificant gates to zero.

What would settle it

Train MobileNetV2 on ImageNet with a fixed $\sigma$ (no scheduling) and measure whether the pruning rate collapses or accuracy drops beyond the reported trade-off; a clear degradation would show that the scheduled-$\sigma$ mechanism is load-bearing.

Watch

Extended reading notes

Core claim

The central claim is that a multiplicative gating layer with activation $1-e^{-g^2}$ (exponential gates) combined with a sparsity regularizer on the gate parameters can drive whole channels to exactly zero during training, while the saturating nature of the proposed bounded-$\ell_1$ penalty lets larger-magnitude gates escape heavy penalization and be optimized primarily by the task loss. This decouples parameter importance from parameter magnitude, which the paper argues improves pruning rates and preserves accuracy compared with standard $\ell_1$ regularization on linear (batch-norm scaling) gates. The paper shows empirically that the exponential gates achieve their best pruning rates at a threshold of exactly zero, meaning removable channels are exactly nulled out, and that the bounded-$\ell_1$ regularizer improves accuracy or pruning rate over plain $\ell_1$ on several models.

Load-bearing premise

The reported pruning-accuracy trade-offs assume that the hand-designed $\sigma$ schedules and per-architecture hyperparameter choices transfer to new architectures and datasets, since the paper provides no principled rule for setting them.

Editorial extensions

If this is right

  • Channel sparsity becomes exact for exponential gates, so pruning thresholds become unnecessary; a zero threshold is nearly optimal.
  • Pruned networks can be deployed without fine-tuning when no batch-norm follows the gate, and with only a few epochs otherwise.
  • The bounded-$\ell_1$ regularizer yields higher pruning rates than plain $\ell_1$ at similar accuracy on MobileNetV2 and ResNet-164, indicating that saturating penalties help retain important channels.
  • Combining exponential gates with $\ell_2$ regularization can improve MobileNetV2 accuracy over standard training while still pruning, suggesting the gating layer acts as a helpful architectural inductive bias.

Reading between the lines

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

  • The same bounded-penalty idea could be applied to other structured units, such as attention heads or transformer blocks, where magnitude-based pruning suffers from the same importance-vs-magnitude entanglement the paper identifies.
  • Because the gates are deterministic and can be merged into the preceding convolution weights, the method may translate to hardware without dedicated sparse kernels, though the paper does not measure latency or energy outcomes.
  • A testable extension: replace the hand-scheduled $\sigma$ with a learned or automatically annealed schedule and check whether the reported pruning/accuracy trade-offs hold outside the tuned hyperparameter ranges.
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

4 major / 4 minor

Summary. This paper proposes two mechanisms for structured channel pruning during training: a trainable exponential gating layer h(g)=1-exp(-g^2) inserted after convolutional layers, and a bounded-l1 regularizer that interpolates between l1 and l0 penalties. The authors argue that exponential gates, combined with l1/l2/bounded-l1 regularization of the gate parameters, drive unimportant channels to exactly zero and thereby remove the need for a carefully chosen pruning threshold. They provide empirical results on LeNet-5/MNIST, DenseNet-40, ResNet-164, and MobileNetV2 on CIFAR-100, plus ResNet-50 and MobileNetV2 on ImageNet, reporting parameter reductions of 30%, 69%, and 75% on CIFAR-100 for ResNet-164, DenseNet-40, and MobileNetV2, and state-of-the-art pruning trade-offs for ResNet-50 on ImageNet. A supplementary section gives the proof of Lemma 1 and detailed training hyperparameters.

Significance. If the empirical claims hold, the paper makes a useful practical contribution: it proposes a simple, trainable group-gating mechanism that can be combined with off-the-shelf regularizers, and a bounded-l1 penalty whose derivative is easy to compute and whose behavior is intuitively appealing. The empirical breadth is a strength: results span several architectures and two datasets of very different scale, and the appendix is transparent about many training details. The mathematical Lemma 1 is simple but correctly proved in the supplement. However, the central exact-zero claim is not rigorously established for the l2-regularized exponential gates used in several headline results, and the comparisons would be stronger with error bars and with baselines trained under identical budgets. These issues are fixable and do not invalidate the overall approach, but they are load-bearing for the paper's main conclusions.

major comments (4)
  1. [Section 4, 'Bounded-l2 for group gating', and Section 5, Fig. 3] The exact-zero claim for exponential gates under l2 regularization is not supported in exact arithmetic. From Eqs. (8)-(9) with R(g)=g^2, if the task-loss gradient for a channel vanishes, the gate obeys dg/dt = -2 lambda2 g, so g(t)=g(0) exp(-2 lambda2 t) > 0 for every finite t; and because h'(0)=0, the same obstruction holds for any smooth regularizer. The threshold-zero pruning reported for l2+exp in Fig. 4 and Table 2 therefore depends either on the discrete optimization dynamics or on numerical underflow, neither of which is described or justified. Please either restrict the exact-zero conclusion to l1/bounded-l1, or define the numerical criterion under which a gate is treated as zero and show that the reported pruning rates are stable under small threshold perturbations and across floating-point precisions.
  2. [Section 5, Tables 1-2 and Fig. 4] Central empirical comparisons are presented without uncertainty measures. Figure 2 reports averages over 3 runs but no error bars, while Table 2 and Fig. 4 report single numbers, and several claimed differences are small in magnitude (e.g., 69.9 vs. 69.54 top-1 accuracy on MobileNetV2, or bounded-l1 vs. l1 on linear gates). Please report standard deviations or confidence intervals over at least three seeds for all headline accuracy/pruning pairs, and state whether the qualitative conclusions are stable across seeds.
  3. [Section 5, Table 2, and Appendix A2] The MobileNetV2 ImageNet comparison is confounded by training budget and by the unvalidated reimplementation of the network-slimming baseline. Table 2 and the appendix state that MobileNetV2 was trained for 100 epochs on ImageNet, in contrast to the standard 400-epoch training, so comparisons with published baselines may reflect training length rather than the pruning method. The paper also states that [24] was reimplemented, but it does not demonstrate that the reimplementation reproduces the original published accuracy/pruning trade-off. Please train all baselines under the identical training schedule and, for reference, also report the original [24] numbers.
  4. [Appendix A2 and Section 4] The reported behavior depends on per-architecture hand-designed sigma schedules and lambda settings (e.g., MobileNetV2 sigma initialized to 2.0 and decayed by 0.99 per epoch; ResNet-164 and DenseNet-40 increase lambda after epoch 120 and use custom sigma decays), yet no selection rule or sensitivity analysis is provided. Because the bounded-l1 regularizer's interpolation between l1 and l0 is controlled by these choices, the claim that the method generalizes to new architectures is currently supported only by anecdotal tuning rather than by a demonstrated robustness to hyperparameter variation.
minor comments (4)
  1. [Section 3, Eq. (4)] The notation in Eq. (4) is garbled in the typeset text; it should read approximately (1/sigma^p) * sum_i |x_i|^p = (||x||_p / sigma)^p. Please fix the formatting.
  2. [Section 3, Eq. (1)] The indicator notation 1_a(b) is unconventional and should be written as 1_{a}(b), with the convention that this equals 1 when a=b and 0 otherwise.
  3. [Figure 2 caption] The caption states that 'two identical markers represent settings with different regularization strengths,' but the markers are not visually distinguished in the figure. Please use different marker styles or annotate the settings directly so the reader can identify them.
  4. [Supplementary, proof of Lemma 1] The proof is correct, but the claim lim_{sigma->0} exp(-|x_i|^p/sigma^p) = 1_0(x_i) should be stated with the convention that the limit at x_i=0 is 1 and at x_i != 0 is 0; this is understood but not explicitly stated.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: the method is empirically evaluated against external baselines, and the only self-citation is a non-load-bearing related-work mention.

full rationale

We walked the claimed derivation chain and found no step in which an output is equivalent to an input by construction. The exponential gating layer is defined in Eq. 8 (h(g)=1-exp(-g^2)) and the regularizers are defined in Eqs. 5, 6, and 9; the reported pruning rates and accuracies are measured outcomes of training with those losses, not quantities fitted from the reported numbers. The paper's pruning protocol uses fixed thresholds (1e-4 for linear gates, zero for exponential gates) stated in Sec. 5, and the empirical comparisons are made against independent external baselines: network slimming from Liu et al. [24], L0 regularization from Louizos et al. [26], and block pruning from Huang and Wang [19]. No parameter is fit to the target pruning rate, and no 'prediction' is renamed from a fitted value. The only author-overlapping citation is [1] (Achterhold et al., which includes author Tim Genewein), used in the related-work survey for weight ternarization; it is not load-bearing for the gating, regularization, or pruning claims. The bounded-Lp norm result in Lemma 1 is a standard limit/Taylor expansion, and the paper attributes the underlying function to Weston et al. [37]. The hand-selected sigma schedules and lambda searches in Sec. A2 are hyperparameter choices, not circular inputs. The mathematical concern that l2-regularized exponential gates cannot reach exact zero in exact arithmetic and may only do so by floating-point underflow is a correctness/numerical reproducibility caveat, not a circularity: the paper's central 'exactly zero' claim is empirical, and its gate definition does not make that claim true by definition. Accordingly, no specific circular reduction can be quoted, and the analysis should not be scored above the minor-self-citation level.

Assumptions & free parameters 5 free parameters · 3 assumptions · 1 invented entities

The central claim depends on several hand-chosen hyperparameters (λ1, λ2, σ schedule, gate initialization, pruning threshold) and on empirical optimization behavior (exact zero convergence). None of these are derived from first principles, so the method's success rests on these choices and observations.

free parameters (5)
  • λ1 (L1/bounded-L1 regularization strength) = e.g., 5e-5 to 1e-4 on ImageNet; 1e-3 to 4e-3 on MNIST; 1e-4 to 5e-4 on CIFAR100
    Chosen by hyperparameter search per architecture and dataset; central to the pruning-accuracy trade-off.
  • λ2 (weight decay) = e.g., 1e-5, 4e-5, 5e-4, 1e-4
    Chosen by hyperparameter search; regularizes all parameters including gates.
  • σ initial value and schedule = e.g., init 2.0; ResNet-164/DenseNet-40: decay 0.02/epoch until 0.2 then 0.99/epoch; MobileNetV2: decay 0.99/epoch
    Hand-designed schedule for bounded-L1; no principled rule, and results depend on it.
  • Gate initialization = g=1.0 for exponential gates; γ=0.5 for linear gates
    Set by hand; affects training dynamics and which channels are pruned.
  • Pruning threshold = 0 for exponential gates; 1e-4 for linear gates
    Selected based on Figure 3 to maximize pruning without accuracy loss.
assumptions (3)
  • standard math Taylor expansion of exp around zero is used to prove Lemma 1 (Eq. 3 in supplementary).
    Standard calculus result; no issue.
  • domain assumption The regularized loss can be optimized by SGD to a state where unimportant gate parameters become exactly zero.
    The paper provides only empirical evidence (Fig. 3) and no convergence proof; this is load-bearing for threshold-zero pruning.
  • domain assumption Pruning channels with zero gates does not significantly alter the network function.
    Empirically observed after fine-tuning for up to three epochs; assumed to hold for all layers and architectures.
invented entities (1)
  • Exponential gating layer
    purpose: Multiplicative channel gate y_k = x_k * (1 - exp(-g_k^2)) that can reach exactly zero when g_k=0, enabling threshold-free pruning.
    Only evaluated within this paper; no external validation, theoretical guarantee, or released implementation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Group Pruning using a Bounded-Lp norm for Group Gating and Regularization." pith.science (2026). https://pith.science/paper/P52PUBAZ

@misc{pith2026190803463,
  author       = {Pith},
  title        = {Pith review of: Group Pruning using a Bounded-Lp norm for Group Gating and Regularization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/P52PUBAZ}},
  note         = {Machine review of arXiv:1908.03463}
}
read the original abstract

Deep neural networks achieve state-of-the-art results on several tasks while increasing in complexity. It has been shown that neural networks can be pruned during training by imposing sparsity inducing regularizers. In this paper, we investigate two techniques for group-wise pruning during training in order to improve network efficiency. We propose a gating factor after every convolutional layer to induce channel level sparsity, encouraging insignificant channels to become exactly zero. Further, we introduce and analyse a bounded variant of the L1 regularizer, which interpolates between L1 and L0-norms to retain performance of the network at higher pruning rates. To underline effectiveness of the proposed methods,we show that the number of parameters of ResNet-164, DenseNet-40 and MobileNetV2 can be reduced down by 30%, 69% and 75% on CIFAR100 respectively without a significant drop in accuracy. We achieve state-of-the-art pruning results for ResNet-50 with higher accuracy on ImageNet. Furthermore, we show that the light weight MobileNetV2 can further be compressed on ImageNet without a significant drop in performance.

Figures

Figures reproduced from arXiv: 1908.03463 by the authors.

Figure 1
Figure 1. Illustration of bounded-`p,0 (b`p,0) norms with p ∈ {1, 2}: Interpolation from `1-norm to 0-norm (left) and from `2-norm to 0-norm (right) with different σ. Bounded-`1 regularizer: It is a common practice to use sparsity inducing `1 penalty to shrink parameters during training. [24] has performed channel-wise pruning by imposing `1 penalty on the scaling factor γ of Batch Normalization (BN) layers that correspond to… view at source ↗
Figure 2
Figure 2. Comparing trade-off between pruning rates and accuracies of different regular￾izers `2, `1 and bounded-`1 with different gates (linear, exponential) at constant and scheduled σ on DenseNet-40, ResNet-164 and MobileNetV2 on CIFAR100. In DenseNet￾40, the scheduled `1 regularizer on exponential gate achieves slightly higher pruning and accuracy rate than the other methods. In ResNet-164, two identical markers represent… view at source ↗
Figure 3
Figure 3. Comparison of pruning rates (top row) and accuracies (bottom row) on CI￾FAR100 over different threshold points between linear gate (blue) and the exponential gate (red), both of which are applied in combination with `1 regularization. Units that do not pass the threshold on the gate values |g| for the linear gates and (1 − e −g 2 ) for exponential gates are pruned. Here, pruning rate of networks with exponential gat… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Comparing our pruning results of ResNet-50 (`2 & `1 on exponential gating layer, `2 + exp and `1 + exp) on ImageNet dataset against the previous methods like ResNet-101(v1) and ResNet-101(v2) from [39], ResNet-32 and ResNet-26 which are obtained from block pruning on R…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 27 canonical work pages

  1. [24]

    In: Computer Vision (ICCV), 2017 IEEE International Conference on

    Liu, Z., Li, J., Shen, Z., Huang, G., Yan, S., Zhang, C.: Learning efficient convo- lutional networks through network slimming. In: Computer Vision (ICCV), 2017 IEEE International Conference on. pp. 2755–2763. IEEE (2017)

    work page 2017

  2. [1]

    ICLR2018 (2018)

    Achterhold, J., Koehler, J.M., Schmeink, A., Genewein, T.: Variational network quantization. ICLR2018 (2018)

    work page 2018

  3. [2]

    Alvarez, J.M., Salzmann, M.: Learning the number of neurons in deep networks. In: Adv. in Neural Info. Process. Syst. (NIPS). pp. 2270–2278 (2016)

    work page 2016

  4. [3]

    Chen, W., Wilson, J., Tyree, S., Weinberger, K., Chen, Y.: Compressing neural networks with the hashing trick. Int. Conf. on Machine Learning (ICML) pp. 2285–2294 (2015)

    work page 2015

  5. [4]

    arXiv:1710.09282 (2017)

    Cheng, Y., Wang, D., Zhou, P., Zhang, T.: A survey of model compression and acceleration for deep neural networks. arXiv:1710.09282 (2017)

    arXiv 2017

  6. [5]

    arXiv:1602.02830 (2016)

    Courbariaux, M., Hubara, I., Soudry, D., El-Yaniv, R., Bengio, Y.: Binarized neural networks: Training deep neural networks with weights and activations constrained to + 1 or -1. arXiv:1602.02830 (2016)

    arXiv 2016

  7. [6]

    Improved Bayesian Compression

    Federici, M., Ullrich, K., Welling, M.: Improved Bayesian compression. arXiv:1711.06494 (2017)

    work page Pith review arXiv 2017

  8. [7]

    arXiv:1803.03635 (2018)

    Frankle, J., Carbin, M.: The lottery ticket hypothesis: Finding small, trainable neural networks. arXiv:1803.03635 (2018)

    arXiv 2018

Show all 41 references
  1. [8]

    arXiv:1806.05975 (2018)

    Ghosh, S., Yao, J., Doshi-Velez, F.: Structured variational learning of Bayesian neural networks with horseshoe priors. arXiv:1806.05975 (2018)

    2018 arXiv

  2. [9]

    arXiv:1412.6115 (2014)

    Gong, Y., Liu, L., Yang, M., Bourdev, L.: Compressing deep convolutional networks using vector quantization. arXiv:1412.6115 (2014)

    2014 arXiv

  3. [10]

    Guo, Y., Yao, A., Chen, Y.: Dynamic network surgery for efficient dnns. Adv. in Neural Info. Process. Syst. (NIPS) pp. 1379–1387 (2016)

    2016

  4. [11]

    IEEE Trans

    Gysel, P., Pimentel, J., Motamedi, M., Ghiasi, S.: Ristretto: A framework for empirical study of resource-efficient inference in convolutional neural networks. IEEE Trans. on Neural Networks and Learning Syst. (2018)

    2018

  5. [12]

    Han, S., Mao, H., Dally, W.J.: Deep compression: Compressing deep neural networks with pruning, trained quantization and Huffman coding. In: Int. Conf. on Learning Representations (ICLR) (2016)

    2016

  6. [13]

    Han, S., Pool, J., Narang, S., Mao, H., Tang, S., Elsen, E., Catanzaro, B., Tran, J., Dally, W.J.: Dsd: Regularizing deep neural networks with dense-sparse-dense training flow. Int. Conf. on Learning Representations (ICLR) (2017)

    2017

  7. [14]

    Han, S., Pool, J., Tran, J., Dally, W.: Learning both weights and connections for efficient neural network. Adv. in Neural Info. Process. Syst. (NIPS) pp. 1135–1143 (2015)

    2015

  8. [15]

    Hanson, S.J., Pratt, L.Y.: Comparing biases for minimal network construction with back-propagation. Adv. in Neural Info. Process. Syst. (NIPS) pp. 177–185 (1989)

    1989

  9. [16]

    In: International Conference on Computer Vision (ICCV)

    He, Y., Zhang, X., Sun, J.: Channel pruning for accelerating very deep neural networks. In: International Conference on Computer Vision (ICCV). vol. 2 (2017)

    2017

  10. [17]

    arXiv:1704.04861 (2017)

    Howard, A.G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., Andreetto, M., Adam, H.: Mobilenets: Efficient convolutional neural networks for mobile vision applications. arXiv:1704.04861 (2017)

    2017 arXiv

  11. [18]

    arXiv:1607.03250 (2016)

    Hu, H., Peng, R., Tai, Y.W., Tang, C.K.: Network trimming: A data-driven neuron pruning approach towards efficient deep architectures. arXiv:1607.03250 (2016)

    2016 arXiv

  12. [19]

    arXiv:1707.01213 (2017)

    Huang, Z., Wang, N.: Data-driven sparse structure selection for deep neural networks. arXiv:1707.01213 (2017)

    2017 arXiv

  13. [20]

    Hubara, I., Courbariaux, M., Soudry, D., El-Yaniv, R., Bengio, Y.: Quantized neural networks: Training neural networks with low precision weights and activations. J. of Machine Learning Research (JMLR) 18(1), 6869–6898 (2017) 14 C. K. Mummadi et al

    2017

  14. [21]

    arXiv:1602.07360 (2016)

    Iandola, F.N., Han, S., Moskewicz, M.W., Ashraf, K., Dally, W.J., Keutzer, K.: Squeezenet: Alexnet-level accuracy with 50x fewer parameters and¡ 0.5 mb model size. arXiv:1602.07360 (2016)

    2016 arXiv

  15. [22]

    arXiv:1505.07765 (2015)

    Karaletsos, T., R¨ atsch, G.: Automatic relevance determination for deep generative models. arXiv:1505.07765 (2015)

    2015 arXiv

  16. [23]

    Li, H., Kadav, A., Durdanovic, I., Samet, H., Graf, H.P.: Pruning filters for efficient convnets. Int. Conf. on Learning Representations (ICLR) (2017)

    2017

  17. [25]

    Advances in Neural Information Processing Systems (2017)

    Louizos, C., Ullrich, K., Welling, M.: Bayesian compression for deep learning. Advances in Neural Information Processing Systems (2017)

    2017

  18. [26]

    ICLR 2018 (2018)

    Louizos, C., Welling, M., Kingma, D.P.: Learning sparse neural networks through l0 regularization. ICLR 2018 (2018)

    2018

  19. [27]

    ICCV2017 (2017)

    Luo, J.H., Wu, J., Lin, W.: Thinet: A filter level pruning method for deep neural network compression. ICCV2017 (2017)

    2017

  20. [28]

    Network: Computation in Neural Systems 6(3), 469–505 (1995)

    MacKay, D.J.: Probable networks and plausible predictions - a review of practical Bayesian methods for supervised neural networks. Network: Computation in Neural Systems 6(3), 469–505 (1995)

    1995

  21. [29]

    ICML 2017 (2017)

    Molchanov, D., Ashukha, A., Vetrov, D.: Variational dropout sparsifies deep neural networks. ICML 2017 (2017)

    2017

  22. [30]

    ICLR2017 (2017)

    Molchanov, P., Tyree, S., Karras, T., Aila, T., Kautz, J.: Pruning convolutional neural networks for resource efficient inference. ICLR2017 (2017)

    2017

  23. [31]

    Neal, R.M.: Bayesian Learning for Neural Networks. Ph.D. thesis, University of Toronto (1995)

    1995

  24. [32]

    arXiv:1705.07283 (2017)

    Neklyudov, K., Molchanov, D., Ashukha, A., Vetrov, D.: Structured Bayesian pruning via log-normal multiplicative noise. arXiv:1705.07283 (2017)

    2017 arXiv

  25. [33]

    In: European Conference on Computer Vision

    Rastegari, M., Ordonez, V., Redmon, J., Farhadi, A.: Xnor-net: Imagenet classi- fication using binary convolutional neural networks. In: European Conference on Computer Vision. pp. 525–542. Springer (2016)

    2016

  26. [34]

    arXiv:1703.09039 (2017)

    Sze, V., Chen, Y.H., Yang, T.J., Emer, J.: Efficient processing of deep neural networks: A tutorial and survey. arXiv:1703.09039 (2017)

    2017 arXiv

  27. [35]

    ICLR 2017 (2017)

    Ullrich, K., Meeds, E., Welling, M.: Soft weight-sharing for neural network com- pression. ICLR 2017 (2017)

    2017

  28. [36]

    In: Advances in Neural Information Processing Systems

    Wen, W., Wu, C., Wang, Y., Chen, Y., Li, H.: Learning structured sparsity in deep neural networks. In: Advances in Neural Information Processing Systems. pp. 2074–2082 (2016)

    2016

  29. [37]

    Weston, J., Elisseeff, A., Sch¨ olkopf, B., Tipping, M.: Use of the zero-norm with linear models and kernel methods. J. of Machine Learning Research (JMLR)

  30. [38]

    arXiv:1802.04680 (2018)

    Wu, S., Li, G., Chen, F., Shi, L.: Training and inference with integers in deep neural networks. arXiv:1802.04680 (2018)

    2018 arXiv

  31. [39]

    arXiv:1802.00124 (2018)

    Ye, J., Lu, X., Lin, Z., Wang, J.Z.: Rethinking the smaller-norm-less-informative assumption in channel pruning of convolution layers. arXiv:1802.00124 (2018)

    2018 arXiv

  32. [40]

    arXiv:1702.03044 (2017)

    Zhou, A., Yao, A., Guo, Y., Xu, L., Chen, Y.: Incremental network quantization: Towards lossless cnns with low-precision weights. arXiv:1702.03044 (2017)

    2017 arXiv

  33. [41]

    In: European Conference on Computer Vision

    Zhou, H., Alvarez, J.M., Porikli, F.: Less is more: Towards compact cnns. In: European Conference on Computer Vision. pp. 662–677. Springer (2016) Group Pruning using a Bounded- 𝓁p norm for Group Gating and Regularization Supplementary material A1 Proof of Lemma 1 (Lemma 1): T...

    2016

Pith tools

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