Pith. sign in

REVIEW 3 major objections 4 minor 98 references

Deep Weight Factorization: Sparse Learning Through the Lens of Artificial Symmetries

T0 review · 3 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Factorizing each weight into $D > 2$ factors lets plain $L_2$ weight decay induce provably equivalent non-smooth $L_{2/D}$ sparse regularization in any neural network.

desk verdict DWF is a genuinely useful extension of shallow factorization with a correct equivalence theorem, but the paper overclaims by implying the algorithm provably reaches the balanced states the theory relies on; the empirical work is strong enough to warrant serious engagement. read the letter →

arxiv 2502.02496 v2 pith:JTKM6TOG submitted 2025-02-04 cs.LG stat.ML

Chris Kolb , Tobias Weber , Bernd Bischl , David Rügamer This is my paper · ORCID
classification cs.LGstat.ML MSC 68T0790C26
keywords deepweightfactorizationsparseregularizationL2/Dquasi-normartificialsymmetryneuralnetworkpruningfactormisalignmentSGDlearningdynamics
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 sparse learning in neural networks can be done without any non-smooth penalty: write every weight as the element-wise product of $D$ factors, train the factors with plain $L_2$ weight decay, then collapse the factors back. It proves that the factorized objective has exactly the same local and global minima as the original network penalized by the non-convex quasi-norm $\|w\|_{2/D}^{2/D}$, which for $D>2$ penalizes small weights more aggressively than $L_1$ and approaches $L_0$. That means standard SGD can induce sparsity directly during training, without proximal solvers or post-hoc pruning. The paper also identifies why naive factor training fails, namely that standard initializations concentrate the product near zero and that small learning rates never reach the balanced-factor regime, and it supplies a truncation-based initialization and a large-learning-rate recipe. On the tested vision benchmarks, depth-3 and depth-4 factorizations keep accuracy at compression ratios where magnitude pruning, SNIP, and SynFlow collapse.

What carries the argument

The load-bearing object is the factor misalignment $M(\omega) = D^{-1} \sum_{d=1}^D \|\omega_d\|_2^2 - \|\omega_1 \odot \cdots \odot \omega_D\|_{2/D}^{2/D}$, which is non-negative and zero exactly at balanced factorizations. The AM-GM inequality gives the lower bound; Lemma 1 shows any local minimizer of the factorized objective must be balanced, and Lemma 4 shows balance is an absorbing state under SGD updates. The equivalence in Theorem 1 then follows by lifting a minimizer of the collapsed problem to a balanced factorization and by constructing balanced factorizations close to any perturbation of a collapsed minimizer. In training dynamics, the paper observes that sparsity emerges precisely as $M(\omega)$ approaches zero, and that large learning rates are needed to drive SGD into this balanced regime.

What would settle it

Train a factorized network while tracking $M(\omega) = D^{-1}\sum_{d=1}^D \|\omega_d\|_2^2 - \|\omega_1 \odot \cdots \odot \omega_D\|_{2/D}^{2/D}$; the theory predicts sparsity should appear only as $M(\omega)$ reaches zero, so a run that achieves high compression with $M(\omega)$ bounded well above zero, or a value of $\lambda$ where the collapsed loss decreases while $M(\omega)$ stays positive, would contradict the claimed equivalence mechanism.

Watch

Extended reading notes

Core claim

The central discovery is Theorem 1: for any network and any $D \geq 2$, the smooth problem $\min_{\omega} L(\omega_1 \odot \cdots \odot \omega_D) + \frac{\lambda}{D} \sum_{d=1}^D \|\omega_d\|_2^2$ and the non-smooth problem $\min_w L(w) + \lambda \|w\|_{2/D}^{2/D}$ share the same global and local minima, with minimizers related by collapse $\hat{w} = \hat{\omega}_1 \odot \cdots \odot \hat{\omega}_D$. The proof rests on the artificial rescaling symmetry of the factorization: for a fixed collapsed weight there are infinitely many factorizations, and the $L_2$ penalty is minimized by balanced factorizations in which all factors share equal magnitude. At balance, the factor penalty equals $\|w\|_{2/D}^{2/D}$, so $L_2$ regularization on factors is a differentiable stand-in for a non-convex sparsity penalty, and deeper factorizations give stronger, more $L_0$-like sparsification.

Load-bearing premise

The method's practical success rests on the assumption that training with the proposed initialization and a large enough learning rate actually drives the factors into the balanced state (all $D$ factors of each weight equal in magnitude) where the theory applies; the paper proves balance is absorbing once reached, not that training converges to it.

Editorial extensions

If this is right

  • Any architecture can be sparsified in principle by training factorized weights with ordinary SGD and weight decay; the collapsed network is a minimizer of an $L_{2/D}$-regularized objective, with no non-smooth solver required.
  • Deeper factorizations ($D=3,4$) act like a stronger, more $L_0$-like penalty, yielding higher compression ratios at the same accuracy than the shallow $D=2$ factorization.
  • Standard initializations provably make factorized networks untrainable: the collapsed product concentrates at zero and activation variance vanishes, so the proposed truncation-and-variance-matching initialization is necessary.
  • Sparsity is not obtained automatically: training must include a large-learning-rate phase, and the paper documents three distinct phases with sparsification onset coinciding with factor balance.
  • DWF can be combined with post-hoc pruning and fine-tuning, further increasing compression by up to roughly three times at matched accuracy.

Reading between the lines

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

  • The factor-misalignment diagnostic suggests a principled stopping rule that the paper does not propose: monitor $M(\omega)$ and treat its approach to zero as the signal that sparsification has begun.
  • Because the equivalence is stated for any differentiable loss, the same factorization trick should extend to structured sparsity by grouping factors, although the paper only treats unstructured sparsity.
  • The observed coincidence of sparsification with balance connects DWF to grokking and stochastic collapse, so the large-learning-rate requirement may be understood as noise-driven symmetry breaking rather than a mere tuning detail.
  • A testable prediction: at fixed $\lambda$, the compression ratio should scale with $D$ through the exponent $2/D$, so measuring sparsity across depths could separate the penalty effect from the initialization effect.
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

3 major / 4 minor

Summary. The paper introduces Deep Weight Factorization (DWF), which factorizes each weight of a neural network into D ≥ 2 factors and trains with L2 weight decay on the factors, collapsing them after training. The main theoretical contribution is Theorem 1, which states that the factorized objective in Eq. (5) and the original objective with the non-convex L2/D quasi-norm penalty in Eq. (4) have the same local and global minima, with minimizers related by collapse. This equivalence relies on Lemma 1 (balanced factorizations are necessary for optimality) and Lemma 3 (balanced factorizations exist locally). The authors also analyze initialization, proposing a variance-matching and truncation scheme (Algorithm 1), identify a large-learning-rate requirement for sparsity, describe three training phases, and empirically benchmark DWF against magnitude pruning, SNIP, and SynFlow on several architectures and datasets.

Significance. If the claims held fully, the paper would be significant: it offers a smooth, SGD-compatible surrogate for a family of non-convex sparsity-inducing penalties, with empirical results that are often substantially better than existing pruning methods at high compression ratios. The theoretical proof of Theorem 1 is clean and sound (AM-GM plus a continuity argument), and the empirical study is extensive, covering multiple architectures, datasets, and factorization depths. The proposed truncation-based initialization is a practical contribution. However, the headline claim that DWF 'provably induces' L2/D regularization in trained networks is not supported: Theorem 1 concerns minimizers of the optimization problems, while the algorithm's convergence to balanced factorizations is only demonstrated empirically. The experimental comparisons also use per-depth/per-architecture tuning for DWF while baselines use fixed standard configurations, and many plots rely on only three seeds. These issues make the practical superiority less definitive than claimed.

major comments (3)
  1. [Section 3 / Theorem 1 / Lemma 4 / Appendix C.4] The central theoretical claim is about the equivalence of minimizers of Eq. (4) and Eq. (5), not about the dynamics of SGD. Lemma 4 shows that the balanced set M(ω)=0 is absorbing under SGD, but it does not establish that the DWF initialization and cosine LR schedule actually drive iterates into this set. The paper's own figures (Figs. 18–20, Appendix F.4) provide empirical evidence of misalignment decay, but the method is sensitive to the initial LR (Fig. 5b, Fig. 11) and to the truncation thresholds in Algorithm 1. Therefore, the statement in the abstract and Section 6 that DWF 'provably induces' L2/D regularization in any network is not justified by the proof; it is a conditional transfer from the optimization problem to the algorithm. I recommend either adding a convergence result (even for a simplified setting) or explicitly qualifying the claim as an empirical finding.
  2. [Section 5.3 / Table 4 / Section G.2] The benchmark comparisons are not controlled for hyperparameter tuning effort. Table 4 and Appendix G.2 show that DWF learning rates are selected per factorization depth and per architecture/dataset (e.g., initial LRs {0.2, 0.5, 0.7} for ResNet-18 CIFAR10; 'we additionally select the best-performing LR ... for each factorization depth, architecture, and dataset'), while the pruning baselines use fixed configurations from prior work. The large improvements reported in Table 1 and Figs. 8–9 could partly reflect this tuning asymmetry. Please add an ablation where DWF uses the same LR for all depths, or where baselines receive comparable tuning, to make the comparison fair.
  3. [Table 2 / Figs. 8–9] Many of the central empirical claims rest on only three random seeds. For example, Fig. 8 reports the 'minimum over three random initializations' and Table 2 reports mean ± std over three seeds; several DWF entries show large variance or collapse, such as LeNet-5 MNIST Depth 2 at 99.5% sparsity (66.88 ± 40.23) and ResNet-18 Tiny ImageNet Depth 4 at 5% tolerance (CR=13, versus CR=67 for Depth 3 in Table 1). These observations indicate that the stability of DWF, especially for D=4, is not yet firmly established. Reporting more seeds (at least 5–10) for the key high-sparsity comparisons, or confidence intervals, would strengthen the claims.
minor comments (4)
  1. [Fig. 21] The axis label contains an encoding artifact ('/uni000000...') that should be fixed.
  2. [Algorithm 2, lines 22–23] The numerical threshold ε_tiny is set to float32 machine epsilon; the sensitivity of the reported sparsities to this threshold is not discussed, though it directly determines what counts as a 'zero' weight.
  3. [Appendix H, Lemma 5] The exact Gaussian product factorization is a nice theoretical aside, but the paper does not compare it to DWF Init in downstream training; a sentence clarifying its role would help.
  4. [Section 6] The conclusion repeats 'provably induces' without referencing the qualification needed in Theorem 1; please align the wording with the actual scope of the proof.

Circularity Check

0 steps flagged · score 2.0 of 10

No load-bearing circularity: Theorem 1 is proved from AM-GM and the balanced-lift Lemma 3; the D>2 sparsity claims are independent of the paper's self-citations.

full rationale

Theorem 1 is not circular: the paper derives it from Lemma 1, where the AM-GM inequality gives D^{-1} sum_d ||omega_d||_2^2 >= ||varpi||_{2/D}^{2/D} with equality iff the factorization is balanced, and from Lemma 3, which constructs a balanced factorization of any nearby collapsed weight. The key identity L_{omega,lambda}(omega) = L_{w,lambda}(K(omega)) + lambda M(omega) holds by the definition of misalignment M, so the minimizer correspondence is obtained from an inequality plus a continuity argument, not from a fitted quantity or from the authors' prior work. The DWF initialization thresholds and learning-rate recommendations in Section 4 are empirical engineering choices; they do not enter Eq. (4)-(5) or Theorem 1, so they are not fitted inputs renamed as predictions. The claimed equivalence is about minimizers of the two optimization problems, and the fact that Lemma 4 only proves balancedness is absorbing under SGD, not that SGD converges to balance, is a limitation in transferring the theorem to actual trajectories; this is a scope gap, not circularity. The only self-citation is Kolb et al. (2023), which appears in related work and as a prior observation that vanilla L1 fails with SGD (Section 5.1); that observation is not load-bearing for the paper's D>2 central claim. Empirical sparsity-accuracy curves in Figs. 1, 8, and 9 are measurements, not predictions derived from a fit. Accordingly, no step of the derivation reduces to its own input.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central theorem relies only on standard inequalities and continuity. Practical performance additionally relies on the hand-designed DWF initialization and LR schedule, which are empirical ingredients, not derived from the theory.

free parameters (4)
  • regularization strength λ = swept over [1e-6, 1e-1] logarithmically
    Controls sparsity level; DWF is evaluated across a sweep, not fit to a target.
  • initial learning rate η0 = per depth/architecture (e.g., 0.15 for LeNet, {0.2,0.5,0.7} for ResNet-18)
    Selected from grid; large LR is required for sparsification.
  • lower truncation threshold ϖ_min = 3e-3
    Hand-set in Algorithm 1; used to avoid dead weights in DWF initialization.
  • factorization depth D = 2, 3, 4 (up to 8 in ablation)
    Model choice; deeper than 4 does not help.
assumptions (5)
  • standard math AM-GM inequality on squared factor magnitudes
    Used in Lemma 1 to show local minima are balanced.
  • standard math Continuity of the collapse map K and Lemma 3 (existence of nearby balanced factorizations)
    Used in Theorem 1 proof to transfer local minima between parameterizations.
  • domain assumption Loss ℓ is continuous and network output is continuous in w
    Implied by Theorem 1 setting; holds for standard networks and losses.
  • ad hoc to paper SGD with weight decay reaches balanced factorizations in practice
    Lemma 4 shows balance is absorbing, not that it is reached; empirical evidence in Figs. 18-20.
  • domain assumption Standard initialization produces iid zero-mean factors with variance σ_l^2
    Used for Lemma 2 to show vanishing of collapsed weights and activations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deep Weight Factorization: Sparse Learning Through the Lens of Artificial Symmetries." pith.science (2026). https://pith.science/paper/JTKM6TOG

@misc{pith2026250202496,
  author       = {Pith},
  title        = {Pith review of: Deep Weight Factorization: Sparse Learning Through the Lens of Artificial Symmetries},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JTKM6TOG}},
  note         = {Machine review of arXiv:2502.02496}
}
abstract

Sparse regularization techniques are well-established in machine learning, yet their application in neural networks remains challenging due to the non-differentiability of penalties like the $L_1$ norm, which is incompatible with stochastic gradient descent. A promising alternative is shallow weight factorization, where weights are decomposed into two factors, allowing for smooth optimization of $L_1$-penalized neural networks by adding differentiable $L_2$ regularization to the factors. In this work, we introduce deep weight factorization, extending previous shallow approaches to more than two factors. We theoretically establish equivalence of our deep factorization with non-convex sparse regularization and analyze its impact on training dynamics and optimization. Due to the limitations posed by standard training practices, we propose a tailored initialization scheme and identify important learning rate requirements necessary for training factorized networks. We demonstrate the effectiveness of our deep weight factorization through experiments on various architectures and datasets, consistently outperforming its shallow counterpart and widely used pruning methods.

Figures

Figures reproduced from arXiv: 2502.02496 by the authors.

Figure 1
Figure 1. Sparsity-accuracy tradeoff using a vanilla [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the proposed method (cf. Algorithm 2). Our approach proceeds by factorizing the neural [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Scalar rescal￾ing symmetry and min￾norm factorizations. While previous works mainly studied rescaling symmetries naturally aris￾ing in, e.g., homogeneous activation functions (Neyshabur et al., 2015; Parhi & Nowak, 2023), weight factorization constitutes an artificial sym￾metry that is independent of L, and by extension also of ℓ(·, ·) and fw(·, x). This applicability to any parametric problem designates artificial … view at source ↗
Figures from the paper (24 more)
Figure 4
Figure 4. Figure 4: DWF initialization strategies. Left: factor densities with variance matching and truncation. Middle: product densities for D = 4 illustrating kurtosis explosion without truncation. Right: sparsity-accuracy curves for different initializations and D, showing the failure…
Figure 5
Figure 5. Figure 5: Failure modes when optimizing factorized neural networks. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Factorized ResNet-18 on CIFAR10 with D = 4. Dashed lines indicate phase transitions. Top: Different LR schedules with same initial LR and λ. Left: cosine LR learns sparse and generalizing solutions. Mid: a const. large LR causes sparsification but no generalization. Ri…
Figure 7
Figure 7. Figure 7: Impact of regularization [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Accuracy vs. sparsity tradeoffs for LeNet architectures on MNIST and replacements of varying [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Accuracy vs. sparsity for larger ResNet and VGG architectures on CIFAR and Tiny ImageNet. [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Allocation of layer-wise sparsity for different methods. SNIP and GMP show catastrophic pruning [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: Sparsity-accuracy tradeoffs for a grid of learning rates, demonstrating the importance of appropri [PITH_FULL_IMAGE:figures/full_fig_p026_11.png]
Figure 12
Figure 12. Figure 12: Experiments on optimal LRs at different amounts of sparsity and different initialization approaches. [PITH_FULL_IMAGE:figures/full_fig_p026_12.png]
Figure 13
Figure 13. Figure 13: Relationship between different regularization strengths and compression ratio (first row), regular [PITH_FULL_IMAGE:figures/full_fig_p027_13.png]
Figure 14
Figure 14. Figure 14: Factorization depths D > 4 empirically do not improve performance but become unstable to train. Sparsity-accuracy curves for LeNet-300-100 on MNIST with increasing LRs shown from left to right. F.2 COMBINED TRAINING AND VALIDATION ACCURACY [PITH_FULL_IMAGE:figures/fu…
Figure 15
Figure 15. Figure 15: Impact of regularization λ on training (top) and validation accuracy (bottom) for VGG-19 on CIFAR100 and D ∈ {2, 3, 4}. The top row shows the training curves for the whole grid of λ values. Bottom row shows validation accuracies without running mean for selected λ. F.…
Figure 16
Figure 16. Figure 16: Impact of regularization λ on compression (top), training, and validation accuracy (bottom) for various architectures and datasets, using D ∈ {2, 3, 4}. rapid and pronounced onset of sparsity across all layers. Different layers exhibit varying rates of sparsification,…
Figure 17
Figure 17. Figure 17: Layer-wise evolution of sparsity (top) and collapsed weight norm (bottom) using D = 3 and increasing regularization λ (left to right) for different architectures and datasets. F.4 EVOLUTION OF MISALIGNMENT AND ONSET OF SPARSITY We investigate the empirical dynamics of…
Figure 18
Figure 18. Figure 18: Evolution of the average layer-wise factor misalignment (dashed) together with layer-wise sparsity [PITH_FULL_IMAGE:figures/full_fig_p031_18.png]
Figure 19
Figure 19. Figure 19: Evolution of factor misalignment M(ω) for VGG-19 on CIFAR100 with increasing λ and factor￾ization depths D ∈ {2, 3, 4} (left to right). The results reveal a clear relationship between the elimination of misalignment and sparsity emer￾gence. The onset of sparsity coinc…
Figure 20
Figure 20. Figure 20: Evolution of loss components and sparsity for ResNet-18 with depth [PITH_FULL_IMAGE:figures/full_fig_p032_20.png]
Figure 21
Figure 21. Figure 21: Additional post-hoc pruning and fine-tuning. ResNet-18 ist first trained with DWF and [PITH_FULL_IMAGE:figures/full_fig_p032_21.png]
Figure 22
Figure 22. Figure 22: Additional experiments applying DWF to WRN-16-8 and ResNet-18. For these experiments, the [PITH_FULL_IMAGE:figures/full_fig_p033_22.png]
Figure 23
Figure 23. Figure 23: Sparsity-accuracy tradeoffs for different depths [PITH_FULL_IMAGE:figures/full_fig_p036_23.png]
Figure 24
Figure 24. Figure 24: Learning dynamics for the root initialization for different learning rate schedules (columns). [PITH_FULL_IMAGE:figures/full_fig_p037_24.png]
Figure 25
Figure 25. Figure 25: Approximation of a Gaussian product for different truncation values of the infinite sum in Lemma 5. [PITH_FULL_IMAGE:figures/full_fig_p039_25.png]
Figure 26
Figure 26. Figure 26: Comparison of wall-clock time per sample for WRN-16-8 ( [PITH_FULL_IMAGE:figures/full_fig_p040_26.png]
Figure 27
Figure 27. Figure 27: Peak GPU memory utilization for WRN-16-8 ( [PITH_FULL_IMAGE:figures/full_fig_p040_27.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

98 extracted references · 52 canonical work pages

  1. [1]

    Sgd with large step sizes learns sparse features

    Maksym Andriushchenko, Aditya Vardhan Varre, Loucas Pillaud-Vivien, and Nicolas Flammarion. Sgd with large step sizes learns sparse features. In International Conference on Machine Learning, pp.\ 903--925. PMLR, 2023

    2023

  2. [2]

    Implicit regularization in deep matrix factorization

    Sanjeev Arora, Nadav Cohen, Wei Hu, and Yuping Luo. Implicit regularization in deep matrix factorization. Advances in Neural Information Processing Systems, 32, 2019

    2019

  3. [3]

    Optimization with sparsity-inducing penalties

    Francis Bach, Rodolphe Jenatton, Julien Mairal, Guillaume Obozinski, et al. Optimization with sparsity-inducing penalties. Foundations and Trends in Machine Learning, 4 0 (1): 0 1--106, 2012

    2012

  4. [4]

    Collapsible linear blocks for super-efficient super resolution

    Kartikeya Bhardwaj, Milos Milosavljevic, Liam O'Neil, Dibakar Gope, Ramon Matas, Alex Chalfin, Naveen Suda, Lingchuan Meng, and Danny Loh. Collapsible linear blocks for super-efficient super resolution. Proceedings of Machine Learning and Systems, 4: 0 529--547, 2022

    2022

  5. [5]

    What is the state of neural network pruning? Proceedings of machine learning and systems, 2: 0 129--146, 2020

    Davis Blalock, Jose Javier Gonzalez Ortiz, Jonathan Frankle, and John Guttag. What is the state of neural network pruning? Proceedings of machine learning and systems, 2: 0 129--146, 2020

    2020

  6. [6]

    Improving network slimming with nonconvex regularization

    Kevin Bui, Fredrick Park, Shuai Zhang, Yingyong Qi, and Jack Xin. Improving network slimming with nonconvex regularization. IEEE Access, 9: 0 115292--115314, 2021

    2021

  7. [7]

    Stochastic collapse: How gradient noise attracts sgd dynamics towards simpler subnetworks

    Feng Chen, Daniel Kunin, Atsushi Yamamura, and Surya Ganguli. Stochastic collapse: How gradient noise attracts sgd dynamics towards simpler subnetworks. Advances in Neural Information Processing Systems, 36, 2024

    2024

  8. [8]

    A survey on deep neural network pruning: Taxonomy, comparison, analysis, and recommendations

    Hongrong Cheng, Miao Zhang, and Javen Qinfeng Shi. A survey on deep neural network pruning: Taxonomy, comparison, analysis, and recommendations. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    2024

Show all 98 references
  1. [9]

    Representation costs of linear neural networks: Analysis and design

    Zhen Dai, Mina Karzand, and Nathan Srebro. Representation costs of linear neural networks: Analysis and design. Advances in Neural Information Processing Systems, 34, 2021

    2021

  2. [10]

    Structured sparsity inducing adaptive optimizers for deep learning

    Tristan Deleu and Yoshua Bengio. Structured sparsity inducing adaptive optimizers for deep learning. arXiv preprint arXiv:2102.03869, 2021

    2021 arXiv

  3. [11]

    Shaving weights with occam's razor: Bayesian sparsification for neural networks using the marginal likelihood

    Rayen Dhahri, Alexander Immer, Bertrand Charpentier, Stephan G \"u nnemann, and Vincent Fortuin. Shaving weights with occam's razor: Bayesian sparsification for neural networks using the marginal likelihood. In Sixth Symposium on Advances in Approximate Bayesian Inference-Non ...

    2024

  4. [12]

    Rigging the lottery: Making all tickets winners

    Utku Evci, Trevor Gale, Jacob Menick, Pablo Samuel Castro, and Erich Elsen. Rigging the lottery: Making all tickets winners. In International conference on machine learning, pp.\ 2943--2952. PMLR, 2020

    2020

  5. [13]

    Variable selection via nonconcave penalized likelihood and its oracle properties

    Jianqing Fan and Runze Li. Variable selection via nonconcave penalized likelihood and its oracle properties. Journal of the American Statistical Association, 96 0 (456): 0 1348--1360, 2001

    2001

  6. [14]

    L. E. Frank and Jerome H Friedman. A statistical view of some chemometrics regression tools. Technometrics, 35 0 (2): 0 109--135, 1993

    1993

  7. [15]

    The lottery ticket hypothesis: Finding sparse, trainable neural networks

    Jonathan Frankle and Michael Carbin. The lottery ticket hypothesis: Finding sparse, trainable neural networks. In International Conference on Learning Representations, 2019

    2019

  8. [16]

    Pruning neural networks at initialization: Why are we missing the mark? In International Conference on Learning Representations, 2020

    Jonathan Frankle, Gintare Karolina Dziugaite, Daniel Roy, and Michael Carbin. Pruning neural networks at initialization: Why are we missing the mark? In International Conference on Learning Representations, 2020

    2020

  9. [17]

    Regularization paths for generalized linear models via coordinate descent

    Jerome Friedman, Trevor Hastie, and Rob Tibshirani. Regularization paths for generalized linear models via coordinate descent. Journal of Statistical Software, 33 0 (1): 0 1, 2010

    2010

  10. [18]

    The state of sparsity in deep neural networks

    Trevor Gale, Erich Elsen, and Sara Hooker. The state of sparsity in deep neural networks. arXiv preprint arXiv:1902.09574, 2019

    1902 arXiv

  11. [19]

    The implicit bias of depth: How incremental learning drives generalization

    Daniel Gissin, Shai Shalev-Shwartz, and Amit Daniely. The implicit bias of depth: How incremental learning drives generalization. In International Conference on Learning Representations, 2019

    2019

  12. [20]

    Hypersparse neural networks: Shifting exploration to exploitation through adaptive regularization

    Patrick Glandorf, Timo Kaiser, and Bodo Rosenhahn. Hypersparse neural networks: Shifting exploration to exploitation through adaptive regularization. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pp.\ 1234--1243, 2023

    2023

  13. [21]

    Understanding the difficulty of training deep feedforward neural networks

    Xavier Glorot and Yoshua Bengio. Understanding the difficulty of training deep feedforward neural networks. In Proceedings of the thirteenth international conference on artificial intelligence and statistics, pp.\ 249--256. JMLR Workshop and Conference Proceedings, 2010

    2010

  14. [22]

    Least absolute shrinkage is equivalent to quadratic penalization

    Yves Grandvalet. Least absolute shrinkage is equivalent to quadratic penalization. In ICANN 98: Proceedings of the 8th International Conference on Artificial Neural Networks, Sk \"o vde, Sweden, 2--4 September 1998 8 , pp.\ 201--206. Springer, 1998

    1998

  15. [23]

    Implicit bias of gradient descent on linear convolutional networks

    Suriya Gunasekar, Jason D Lee, Daniel Soudry, and Nati Srebro. Implicit bias of gradient descent on linear convolutional networks. Advances in Neural Information Processing Systems, 31, 2018

    2018

  16. [24]

    Expandnets: Linear over-parameterization to train compact convolutional networks

    Shuxuan Guo, Jose M Alvarez, and Mathieu Salzmann. Expandnets: Linear over-parameterization to train compact convolutional networks. Advances in Neural Information Processing Systems, 33: 0 1298--1310, 2020

    2020

  17. [25]

    Learning both weights and connections for efficient neural network

    Song Han, Jeff Pool, John Tran, and William Dally. Learning both weights and connections for efficient neural network. Advances in neural information processing systems, 28, 2015

    2015

  18. [26]

    Matrix completion and low-rank svd via fast alternating least squares

    Trevor Hastie, Rahul Mazumder, Jason D Lee, and Reza Zadeh. Matrix completion and low-rank svd via fast alternating least squares. The Journal of Machine Learning Research, 16 0 (1): 0 3367--3402, 2015

    2015

  19. [27]

    Delving deep into rectifiers: Surpassing human-level performance on imagenet classification

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Delving deep into rectifiers: Surpassing human-level performance on imagenet classification. In Proceedings of the IEEE International Conference on Computer Vision, pp.\ 1026--1034, 2015

    2015

  20. [28]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp.\ 770--778, 2016

    2016

  21. [29]

    Structured pruning for deep convolutional neural networks: A survey

    Yang He and Lingao Xiao. Structured pruning for deep convolutional neural networks: A survey. IEEE transactions on pattern analysis and machine intelligence, 2023

    2023

  22. [30]

    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, pp.\ 1389--1397, 2017

    2017

  23. [31]

    Sparsity in deep learning: Pruning and growth for efficient inference and training in neural networks

    Torsten Hoefler, Dan Alistarh, Tal Ben-Nun, Nikoli Dryden, and Alexandra Peste. Sparsity in deep learning: Pruning and growth for efficient inference and training in neural networks. Journal of Machine Learning Research, 22 0 (241): 0 1--124, 2021

    2021

  24. [32]

    Lasso, fractional norm and structured sparse estimation using a hadamard product parametrization

    Peter D Hoff. Lasso, fractional norm and structured sparse estimation using a hadamard product parametrization. Computational Statistics & Data Analysis, 115: 0 186--198, 2017

    2017

  25. [33]

    Group sparse optimization via _ p,q regularization

    Yaohua Hu, Chong Li, Kaiwen Meng, Jing Qin, and Xiaoqi Yang. Group sparse optimization via _ p,q regularization. The Journal of Machine Learning Research, 18 0 (1): 0 960--1011, 2017

    2017

  26. [34]

    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 International Conference on Machine Learning, pp.\ 448--456. pmlr, 2015

    2015

  27. [35]

    Implicit bias of large depth networks: a notion of rank for nonlinear functions

    Arthur Jacot. Implicit bias of large depth networks: a notion of rank for nonlinear functions. In The Eleventh International Conference on Learning Representations, 2023

    2023

  28. [36]

    Implicit rank-minimizing autoencoder

    Li Jing, Jure Zbontar, et al. Implicit rank-minimizing autoencoder. Advances in Neural Information Processing Systems, 33: 0 14736--14746, 2020

    2020

  29. [37]

    Dynamic sparse training: Find efficient sparse network from scratch with trainable masked layers

    LIU Junjie, XU Zhe, SHI Runbin, Ray CC CHEUNG, and KH Hayden. Dynamic sparse training: Find efficient sparse network from scratch with trainable masked layers. In 8th International Conference on Learning Representations (ICLR 2020). International Conference on Learning Represe...

    2020

  30. [38]

    u ller, Bernd Bischl, and David R \

    Chris Kolb, Christian L M \"u ller, Bernd Bischl, and David R \"u gamer. Smoothing the edges: a general framework for smooth optimization in sparse regularization using hadamard overparametrization. arXiv preprint arXiv:2307.03571, 2023

    2023 arXiv

  31. [39]

    Neural mechanics: Symmetry and broken conservation laws in deep learning dynamics

    Daniel Kunin, Javier Sagastuy-Brena, Surya Ganguli, Daniel LK Yamins, and Hidenori Tanaka. Neural mechanics: Symmetry and broken conservation laws in deep learning dynamics. In International Conference on Learning Representations, 2020

    2020

  32. [40]

    Soft threshold weight reparameterization for learnable sparsity

    Aditya Kusupati, Vivek Ramanujan, Raghav Somani, Mitchell Wortsman, Prateek Jain, Sham Kakade, and Ali Farhadi. Soft threshold weight reparameterization for learnable sparsity. In International Conference on Machine Learning, pp.\ 5544--5555. PMLR, 2020

    2020

  33. [41]

    Training invariances and the low-rank phenomenon: beyond linear networks

    Thien Le and Stefanie Jegelka. Training invariances and the low-rank phenomenon: beyond linear networks. In International Conference on Learning Representations, 2022

    2022

  34. [42]

    Optimal brain damage

    Yann LeCun, John Denker, and Sara Solla. Optimal brain damage. Advances in neural information processing systems, 2, 1989

    1989

  35. [43]

    Gradient-based learning applied to document recognition

    Yann LeCun, L \'e on Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86 0 (11): 0 2278--2324, 1998

    1998

  36. [44]

    Efficient backprop

    Yann LeCun, L \'e on Bottou, Genevieve B Orr, and Klaus-Robert M \"u ller. Efficient backprop. In Neural networks: Tricks of the trade, pp.\ 9--50. Springer, 2002

    2002

  37. [45]

    Snip: single-shot network pruning based on connection sensitivity

    N Lee, T Ajanthan, and P Torr. Snip: single-shot network pruning based on connection sensitivity. In International Conference on Learning Representations. Open Review, 2019

    2019

  38. [46]

    Pruning filters for efficient convnets

    Hao Li, Asim Kadav, Igor Durdanovic, Hanan Samet, and Hans Peter Graf. Pruning filters for efficient convnets. In International Conference on Learning Representations, 2022

    2022

  39. [47]

    Implicit sparse regularization: The impact of depth and early stopping

    Jiangyuan Li, Thanh Nguyen, Chinmay Hegde, and Ka Wai Wong. Implicit sparse regularization: The impact of depth and early stopping. Advances in Neural Information Processing Systems, 34, 2021

    2021

  40. [48]

    Improving adaptivity via over-parameterization in sequence models

    Yicheng Li and Qian Lin. Improving adaptivity via over-parameterization in sequence models. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

    2024

  41. [49]

    Reconciling modern deep learning with traditional optimization analyses: The intrinsic learning rate

    Zhiyuan Li, Kaifeng Lyu, and Sanjeev Arora. Reconciling modern deep learning with traditional optimization analyses: The intrinsic learning rate. Advances in Neural Information Processing Systems, 33: 0 14544--14555, 2020

    2020

  42. [50]

    Learning efficient convolutional networks through network slimming

    Zhuang Liu, Jianguo Li, Zhiqiang Shen, Gao Huang, Shoumeng Yan, and Changshui Zhang. Learning efficient convolutional networks through network slimming. In Proceedings of the IEEE international conference on computer vision, pp.\ 2736--2744, 2017

    2017

  43. [51]

    Omnigrok: Grokking beyond algorithmic data

    Ziming Liu, Eric J Michaud, and Max Tegmark. Omnigrok: Grokking beyond algorithmic data. In The Eleventh International Conference on Learning Representations, 2023

    2023

  44. [52]

    Sgdr: Stochastic gradient descent with warm restarts

    Ilya Loshchilov and Frank Hutter. Sgdr: Stochastic gradient descent with warm restarts. In International Conference on Learning Representations, 2022

    2022

  45. [53]

    Christos Louizos, Max Welling, and Diederik P. Kingma. Learning sparse neural networks through l_0 regularization. In International Conference on Learning Representations, 2018

    2018

  46. [54]

    Spectral regularization algorithms for learning large incomplete matrices

    Rahul Mazumder, Trevor Hastie, and Robert Tibshirani. Spectral regularization algorithms for learning large incomplete matrices. The Journal of Machine Learning Research, 11: 0 2287--2322, 2010

    2010

  47. [55]

    High-dimensional graphs and variable selection with the lasso

    Nicolai Meinshausen and Peter B \"u hlmann. High-dimensional graphs and variable selection with the lasso. The Annals of Statistics, 34 0 (3): 0 1436--1462, 2006

    2006

  48. [56]

    Implicit bias of the step size in linear diagonal neural networks

    Mor Shpigel Nacson, Kavya Ravichandran, Nathan Srebro, and Daniel Soudry. Implicit bias of the step size in linear diagonal neural networks. In International Conference on Machine Learning, pp.\ 16270--16295. PMLR, 2022

    2022

  49. [57]

    In search of the real inductive bias: On the role of implicit regularization in deep learning

    Behnam Neyshabur, Ryota Tomioka, and Nathan Srebro. In search of the real inductive bias: On the role of implicit regularization in deep learning. In ICLR (Workshop), 2015

    2015

  50. [58]

    Decoupled weight decay for any p norm

    Nadav Joseph Outmezguine and Noam Levi. Decoupled weight decay for any p norm. arXiv preprint arXiv:2404.10824, 2024

    2024 arXiv

  51. [59]

    Kurdyka-- ojasiewicz exponent via hadamard parametrization

    Wenqing Ouyang, Yuncheng Liu, Ting Kei Pong, and Hao Wang. Kurdyka-- ojasiewicz exponent via hadamard parametrization. SIAM Journal on Optimization, 35 0 (1): 0 62--91, 2025

    2025

  52. [60]

    Deep learning meets sparse regularization: A signal processing perspective

    Rahul Parhi and Robert D Nowak. Deep learning meets sparse regularization: A signal processing perspective. IEEE Signal Processing Magazine, 40 0 (6): 0 63--74, 2023

    2023

  53. [61]

    Implicit bias of sgd for diagonal linear networks: a provable benefit of stochasticity

    Scott Pesme, Loucas Pillaud-Vivien, and Nicolas Flammarion. Implicit bias of sgd for diagonal linear networks: a provable benefit of stochasticity. Advances in Neural Information Processing Systems, 34: 0 29218--29230, 2021

    2021

  54. [62]

    The exp-normal distribution is infinitely divisible

    Iosif Pinelis. The exp-normal distribution is infinitely divisible. arXiv preprint arXiv:1803.09838, 2018

    2018 arXiv

  55. [63]

    Smooth bilevel programming for sparse regularization

    Clarice Poon and Gabriel Peyr \'e . Smooth bilevel programming for sparse regularization. Advances in Neural Information Processing Systems, 34, 2021

    2021

  56. [64]

    Smooth over-parameterized solvers for non-smooth structured optimization

    Clarice Poon and Gabriel Peyr \'e . Smooth over-parameterized solvers for non-smooth structured optimization. Mathematical Programming, pp.\ 1--56, 2023

    2023

  57. [65]

    Grokking: Generalization beyond overfitting on small algorithmic datasets

    Alethea Power, Yuri Burda, Harri Edwards, Igor Babuschkin, and Vedant Misra. Grokking: Generalization beyond overfitting on small algorithmic datasets. arXiv preprint arXiv:2201.02177, 2022

    2022 arXiv

  58. [66]

    Winning the lottery with continuous sparsification

    Pedro Savarese, Hugo Silva, and Michael Maire. Winning the lottery with continuous sparsification. Advances in neural information processing systems, 33: 0 11380--11390, 2020

    2020

  59. [67]

    Group sparse regularization for deep neural networks

    Simone Scardapane, Danilo Comminiello, Amir Hussain, and Aurelio Uncini. Group sparse regularization for deep neural networks. Neurocomputing, 241: 0 81--89, 2017

    2017

  60. [68]

    Powerpropagation: A sparsity inducing weight reparameterisation

    Jonathan Schwarz, Siddhant Jayakumar, Razvan Pascanu, Peter Latham, and Yee Teh. Powerpropagation: A sparsity inducing weight reparameterisation. Advances in Neural Information Processing Systems, 34, 2021

    2021

  61. [69]

    A unified scalable equivalent formulation for schatten quasi-norms

    Fanhua Shang, Yuanyuan Liu, Fanjie Shang, Hongying Liu, Lin Kong, and Licheng Jiao. A unified scalable equivalent formulation for schatten quasi-norms. Mathematics, 8 0 (8): 0 1325, 2020

    2020

  62. [70]

    Very deep convolutional networks for large-scale image recognition

    Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556, 2014

    2014 arXiv

  63. [71]

    Geometry of the loss landscape in overparameterized neural networks: Symmetries and invariances

    Berfin Simsek, Fran c ois Ged, Arthur Jacot, Francesco Spadaro, Cl \'e ment Hongler, Wulfram Gerstner, and Johanni Brea. Geometry of the loss landscape in overparameterized neural networks: Symmetries and invariances. In International Conference on Machine Learning, pp.\ 9722-...

    2021

  64. [72]

    Nathan Srebro, Jason D. M. Rennie, and Tommi S. Jaakkola. Maximum-margin matrix factorization. In Proceedings of the 17th International Conference on Neural Information Processing Systems, NIPS'04, pp.\ 1329–1336, Cambridge, MA, USA, 2004. MIT Press

    2004

  65. [73]

    Pruning neural networks without any data by iteratively conserving synaptic flow

    Hidenori Tanaka, Daniel Kunin, Daniel L Yamins, and Surya Ganguli. Pruning neural networks without any data by iteratively conserving synaptic flow. Advances in neural information processing systems, 33: 0 6377--6389, 2020

    2020

  66. [74]

    A comprehensive survey on regularization strategies in machine learning

    Yingjie Tian and Yuqi Zhang. A comprehensive survey on regularization strategies in machine learning. Information Fusion, 80: 0 146--166, 2022

    2022

  67. [75]

    Regression shrinkage and selection via the lasso

    Robert Tibshirani. Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society: Series B (Methodological), 58 0 (1): 0 267--288, 1996

    1996

  68. [76]

    Equivalences between sparse models and neural networks

    Ryan Tibshirani. Equivalences between sparse models and neural networks. Working Notes, 2021

    2021

  69. [77]

    Implicit regularization for optimal sparse recovery

    Tomas Vaskevicius, Varun Kanade, and Patrick Rebeschini. Implicit regularization for optimal sparse recovery. Advances in Neural Information Processing Systems, 32, 2019

    2019

  70. [78]

    Picking winning tickets before training by preserving gradient flow

    Chaoqi Wang, Guodong Zhang, and Roger Grosse. Picking winning tickets before training by preserving gradient flow. In International Conference on Learning Representations, 2020

    2020

  71. [79]

    Why is the state of neural network pruning so confusing? on the fairness, comparison setup, and trainability in network pruning

    Huan Wang, Can Qin, Yue Bai, and Yun Fu. Why is the state of neural network pruning so confusing? on the fairness, comparison setup, and trainability in network pruning. arXiv preprint arXiv:2301.05219, 2023

    2023 arXiv

  72. [80]

    Random weight factorization improves the training of continuous neural representations

    Sifan Wang, Hanwen Wang, Jacob H Seidman, and Paris Perdikaris. Random weight factorization improves the training of continuous neural representations. arXiv preprint arXiv:2210.01274, 2022

    2022 arXiv

  73. [81]

    Learning structured sparsity in deep neural networks

    Wei Wen, Chunpeng Wu, Yandan Wang, Yiran Chen, and Hai Li. Learning structured sparsity in deep neural networks. In Advances in Neural Information Processing Systems, volume 29, 2016

    2016

  74. [82]

    Kernel and rich regimes in overparametrized models

    Blake Woodworth, Suriya Gunasekar, Jason D Lee, Edward Moroshko, Pedro Savarese, Itay Golan, Daniel Soudry, and Nathan Srebro. Kernel and rich regimes in overparametrized models. In Conference on Learning Theory, pp.\ 3635--3673. PMLR, 2020

    2020

  75. [83]

    L1/2 regularization

    Zongben Xu, Hai Zhang, Yao Wang, XiangYu Chang, and Yong Liang. L1/2 regularization. Science China Information Sciences, 53 0 (6): 0 1159--1169, 2010

    2010

  76. [84]

    Proxsgd: Training structured neural networks under regularization and constraints

    Yang Yang, Yaxiong Yuan, Avraam Chatzimichailidis, Ruud JG van Sloun, Lei Lei, and Symeon Chatzinotas. Proxsgd: Training structured neural networks under regularization and constraints. In International Conference on Learning Representations (ICLR) 2020, 2020

    2020

  77. [85]

    92.45\ https://torch.ch/blog/2015/07/30/cifar.html, 2015

    Sergey Zagoruyko. 92.45\ https://torch.ch/blog/2015/07/30/cifar.html, 2015. Torch Blog

    2015

  78. [86]

    Wide residual networks

    Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. In British Machine Vision Conference 2016. British Machine Vision Association, 2016

    2016

  79. [87]

    Nearly unbiased variable selection under minimax concave penalty

    Cun-Hui Zhang. Nearly unbiased variable selection under minimax concave penalty. The Annals of Statistics, 38 0 (2): 0 894--942, 2010

    2010

  80. [88]

    The sparsity and bias of the lasso selection in high-dimensional linear regression

    Cun-Hui Zhang and Jian Huang. The sparsity and bias of the lasso selection in high-dimensional linear regression. The Annals of Statistics, 36 0 (4): 0 1567--1594, 2008

    2008

  81. [89]

    How sparse can we prune a deep network: A fundamental limit perspective

    Qiaozhe Zhang, Ruijie ZHANG, Jun Sun, and Yingzhuang Liu. How sparse can we prune a deep network: A fundamental limit perspective. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

    2024

  82. [90]

    High-dimensional linear regression via implicit regularization

    Peng Zhao, Yun Yang, and Qiao-Chu He. High-dimensional linear regression via implicit regularization. Biometrika, 2022

    2022

  83. [91]

    Effective sparsification of neural networks with global sparsity constraint

    Xiao Zhou, Weizhong Zhang, Hang Xu, and Tong Zhang. Effective sparsification of neural networks with global sparsity constraint. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp.\ 3599--3608, 2021

    2021

  84. [92]

    Symmetry induces structure and constraint of learning

    Liu Ziyin. Symmetry induces structure and constraint of learning. In Forty-first International Conference on Machine Learning, 2023

    2023

  85. [93]

    spred: Solving l1 penalty with sgd

    Liu Ziyin and Zihao Wang. spred: Solving l1 penalty with sgd. In International Conference on Machine Learning, pp.\ 43407--43422. PMLR, 2023

    2023

  86. [94]

    The probabilistic stability of stochastic gradient descent

    Liu Ziyin, Botao Li, Tomer Galanti, and Masahito Ueda. The probabilistic stability of stochastic gradient descent. arXiv preprint arXiv:2303.13093, 2023

    2023 arXiv

  87. [95]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

  88. [96]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should ...

  89. [97]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@firs...

  90. [98]

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibset...

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.