Pith. sign in

REVIEW 3 major objections 4 minor 38 references

Refining the Structure of Neural Networks Using Matrix Conditioning

T0 review · 3 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 Matrix conditioning of partially trained weights identifies which layers carry redundant neurons, so a feed-forward network can be sized without brute-force search.

desk verdict A cheap, clearly specified architecture-search heuristic, but the load-bearing condition-number criterion is not invariant under neuron rescaling, and the manuscript carries an unfinished editorial note. read the letter →

arxiv 1908.02400 v1 pith:3TIN7EKX submitted 2019-08-06 cs.LG cs.NAmath.NAstat.ML

Roozbeh Yousefzadeh , Dianne P O'Leary This is my paper · ORCID
classification cs.LGcs.NAmath.NAstat.ML MSC 68T0562M4565F30
keywords neuralarchitecturedesignmatrixconditioningconditionnumbersingularvaluedecompositionpivotedQRneuronpruningfeed-forwardnetworksmodelcompression
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 claims that the condition number of each layer's stacked weight-bias matrix, inspected after only one to three training epochs, reveals which layers have more neurons than they need. If one layer's matrix is much worse conditioned than the others, or nearly rank-deficient, the layer is over-proportioned; removing neurons until all layers have condition numbers below a common threshold yields the right relative layer sizes. Scaling that skeleton up or down by uniform factors and choosing the scale with the lowest validation-plus-generalization error then produces a compact network. The authors demonstrate the pipeline on MNIST and the Adult Income dataset, reaching accuracies comparable to the best reported models, and add a QR-based procedure that removes redundant neurons from already-trained networks, sometimes improving accuracy after a short retrain.

What carries the argument

The central object is the stacked weight-bias matrix $\widehat{W}^{(i)}$ and its 2-norm condition number $\kappa(\widehat{W}^{(i)}) = \sigma_{\max}/\sigma_{\min}$, computed with SVD, rank-revealing QR, or pivoted QR decomposition. Algorithm 3.1 repeatedly partially trains the network for a small number of epochs $\eta$ and deletes neurons from every layer whose condition number exceeds the threshold $\tau$, using the number of small singular values (or small diagonal entries of $R$ in a QR variant) to decide how many neurons to remove. Algorithm 3.2 then scales all hidden layers by factors $\beta_j$, partially trains each candidate a few times, and picks the size minimizing $2\hat{\epsilon}_{v} - \hat{\epsilon}_{tr}$, a combination of validation error and estimated generalization gap. Algorithm 4.1 applies pivoted QR to a fully trained network so the permutation matrix identifies exactly which columns and neurons are redundant and can be removed while keeping the rest of the trained network intact.

What would settle it

Run Algorithm 3.1 on a network in which one hidden layer is deliberately made far too wide and a later hidden layer is deliberately made too narrow, then fully train the pruned network. If the intentionally narrow layer ends up with a higher condition number after one epoch and the algorithm therefore removes neurons from it, while the over-wide layer's condition number stays below $\tau$, the resulting network will be less accurate than the same starting network pruned by removing the same number of neurons from the over-wide layer, refuting the claim that condition number tracks over-proportioning.

Watch

Extended reading notes

Core claim

The paper's central claim is that matrix conditioning of partially trained parameters can drive architecture design. Let $\widehat{W}^{(i)}$ be the weight matrix of layer $i$ with the bias vector appended as a row, and let $\kappa(\widehat{W}^{(i)})$ be its 2-norm condition number. The claim is that if $\kappa(\widehat{W}^{(i)})$ is large relative to other layers, or if the matrix is close to rank deficient, layer $i$ has an over-proportioned number of neurons; trimming those layers until all condition numbers fall below a threshold $\tau$, then scaling the whole network while preserving layer proportions, finds a small network with high accuracy. The paper also claims that pivoted QR decomposition of a fully trained network identifies precisely which neurons to remove, and that squeezing an oversized trained network in this way can reduce overfitting and even improve test accuracy after retraining. These claims are supported on MNIST and Adult Income, where the automatically designed networks reach about 98.7% and 86.1% test accuracy.

Load-bearing premise

The whole method rests on trusting that after one to three epochs of training, the condition number of a layer's weight matrix tells you whether that layer has too many neurons; if early-training conditioning does not track each layer's eventual value to the network, the recipe for pruning and scaling collapses.

Editorial extensions

If this is right

  • A practitioner with a new feed-forward task and a chosen depth can obtain a near-appropriate per-layer neuron distribution from a few epochs of partial training on a rough starting network, without training many full models.
  • Because Algorithm 3.2 only changes overall scale and not proportions, the expensive search over network size reduces to evaluating a handful of scaled models, and the selected model should be close to the best achievable among all scales.
  • For an already-trained oversized network, squeezing with pivoted QR removes redundant neurons cheaply, and retraining the squeezed model can match or improve the original accuracy while cutting parameter count.
  • The methods do not aim to set benchmark records; their value is reaching modest target accuracy with a compact network and avoiding overfitting, so modest applications with limited compute can afford a tailored architecture.

Reading between the lines

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

  • An unstated consequence is that the same two-step recipe could serve as a cheap first pass in neural architecture search: use the condition-number signal to shrink the candidate space of depths and widths before any expensive search begins, since the paper's evidence suggests the signal is visible within a single epoch.
  • The threshold $\tau$ is effectively a free parameter tuned by looking at the spread of condition numbers; a natural extension the authors do not pursue is to set $\tau$ automatically as a function of that spread, for example as a quantile, so the pipeline becomes fully parameter-free.
  • Because the paper only tests two fully connected datasets, the most direct test of the mechanism on other architectures would be to apply the same partial-training condition-number check inside convolutional or residual networks and see whether the flagged layers align with layers that are actually over-wide.
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. This paper proposes three algorithms for automatically determining the number of neurons in each layer of a feed-forward neural network. Algorithm 3.1 removes neurons from layers whose stacked weight-bias matrix has a condition number exceeding a threshold τ after short partial training, with the aim of equalizing condition numbers across layers. Algorithm 3.2 scales the resulting layer proportions up or down by factors β, partially trains each candidate q times, and selects the model minimizing 2ε_v − ε_tr. Algorithm 4.1 squeezes an already trained network by using pivoted QR on the stacked matrices to remove columns and associated neurons below a threshold. Experiments on MNIST (using 200 Haar-wavelet features) and the Adult Income dataset report compact networks with accuracies of 98.68% and 86.05%, and the selected β is confirmed as the best in the candidate pool after full training on both datasets.

Significance. If the conditioning heuristic were reliable, the paper would offer a very cheap alternative to neural architecture search: the matrix decompositions are inexpensive relative to training, Algorithms 3.1 and 3.2 avoid training many full models, and Algorithm 4.1 gives a simple post-hoc squeezing procedure. The internal verification is a real strength: the β selected by the validation-based score is indeed the best of the eight candidates on the test set for both datasets, and the algorithm descriptions are precise enough to be reproduced. However, the central premise is an empirically motivated heuristic whose main risk—dependence on arbitrary neuron scaling—is not examined, and the evaluation is too narrow to establish the general claim. The paper would be substantially strengthened by a scale-invariant conditioning measure and a broader evaluation with baselines.

major comments (3)
  1. [Section 3 (first insight; Algorithms 3.1 and 4.1)] The central premise that a high condition number of the stacked matrix Ŵ^(i) indicates an over-proportioned layer is not invariant under a symmetry of the network. For a ReLU (or leaky ReLU) network, choose any hidden neuron j in layer i and any c > 0; multiplying column j and bias entry j of Ŵ^(i) by c and dividing row j of W^(i+1) by c leaves the network function unchanged, but changes the singular values of Ŵ^(i) and can make κ(Ŵ^(i)) arbitrarily large. Because Algorithm 4.1 uses pivoted QR, which is column-norm dependent, the same trained function can receive opposite squeezing recommendations depending on the arbitrary scale of a single neuron. The paper does not address this; Section 5.5 only discusses η and τ. Please either define a scale-invariant conditioning measure (e.g., column equilibration of Ŵ^(i) before computing κ) or explicitly restrict the claims to a canonical parameterization and provide evidence that the effect does not occur under the used initialization and optimizer.
  2. [Section 5.5 (hyperparameter selection) and Tables 3, 5] The method's free parameters τ and η are calibrated on the same two datasets on which the method is evaluated. Section 5.5 states that τ should be chosen from the observed variance of condition numbers and that η is chosen after an initial experiment on the user's dataset; the reported experiments use τ = 25 and 40 for MNIST and Adult, respectively, and η = 1 and 3. The 'Verifying the results' subsections (Tables 3 and 5) only confirm that the β selected by Algorithm 3.2 is the best in the pool after full training on the same datasets; this is an internal consistency check, not an out-of-sample evaluation of the design procedure. Since the motivation is to handle unfamiliar datasets, an evaluation that holds out the τ/η selection rule (or tests on additional datasets without recalibration) is needed to support the generalization claim.
  3. [Sections 5.1–5.4 (evaluation scope)] The empirical support for the general claims is limited to two datasets, and the MNIST setup is nonstandard: inputs are 200 Haar wavelet coefficients, the network is a plain feed-forward net with no regularization, and accuracy is 98.68%. There are no comparisons with standard architecture-search or pruning baselines under the same budget, and reported accuracies are single runs without error bars (except the q = 5 averaging in Algorithm 3.2). The conclusion in Section 6 that these are 'a complete set of inexpensive and approachable tools' for arbitrary datasets goes beyond the evidence. Additional datasets, baselines, and variance estimates would considerably strengthen the paper.
minor comments (4)
  1. [Section 5.5] The text 'What does "adaptive" mean in the next sentence? I meant τ is the mean of condition numbers obtained at each iteration. I changed the sentence.' appears to be an editing note accidentally left in the manuscript; please remove or rewrite the passage.
  2. [Footnote 1 (author affiliation)] The affiliation footnote reads 'Department of of Computer Science', with a duplicated 'of'; please correct the typo.
  3. [Section 1.4] The phrase 'Our three algoritms' contains a typo; it should be 'algorithms'.
  4. [Section 5.5] In the sentence about starting with a modest network, 'contacting it with Algorithm 3.2' should likely be 'contracting it with Algorithm 3.2' (or 'connecting').

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the conditioning criterion is a stated heuristic, and the paper's accuracy claims are validated empirically rather than derived from the criterion by construction.

full rationale

The paper's central premise—that high condition numbers in partially trained stacked weight-bias matrices indicate an over-proportioned layer—is introduced explicitly as an 'insight' and a heuristic, not as a theorem derived from earlier results. Algorithm 3.1 prunes layers until all condition numbers fall below τ, so the final condition numbers being below τ are a termination condition rather than an independent prediction; the paper does not rest its central claim on that observation. The actual evidence for the method is external: final networks are fully trained and evaluated on MNIST and Adult test sets, and the 'Verifying the results' subsections fully train all candidate β-scaled models and confirm that the model selected by Algorithm 3.2 indeed has the best test accuracy. Algorithm 4.1 similarly identifies neurons via pivoted QR and then assesses the squeezed network by test accuracy and retraining, so the pruning decisions are checked against an independent objective rather than being defined as correct by the condition-number criterion. The self-citation [34] appears only in an aside about feature engineering and is not load-bearing for any algorithmic claim. The paper's own Section 5.5 limitation—'we cannot guarantee that one (or half) epoch will be adequate for all datasets'—further confirms that the conditioning heuristic is presented as an empirical assumption with acknowledged boundaries. Concerns about in-sample hyperparameter choice, lack of external benchmarks, and non-invariance under neuron rescaling are substantive correctness risks, but they are not instances of self-definition, fitted-inputs-renamed-as-predictions, or load-bearing self-citation under the criteria used here.

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

The central method rests on a stated heuristic rather than a theorem. The biggest free parameters are tau and eta, which are calibrated on the two datasets. The axioms capture the unproven assumptions that condition numbers expose neuron redundancy and that the validation-score heuristic selects the best scale. No new entities are invented.

free parameters (4)
  • tau (condition number threshold) = 20-50 (25-40 suggested)
    Used in Algorithms 3.1 and 4.1 to decide when a layer has excessive neurons; the paper states that values between 20 and 50 lead to similar networks, and that the best tau correlates with small variance of condition numbers. It is calibrated on the two datasets.
  • eta (partial-training epochs) = 1 for MNIST, 3 for Adult
    Number of epochs before condition numbers are inspected; the paper says even half an epoch captures the conditioning but admits there is no guarantee for all datasets.
  • beta scaling range = 1.0 to 2.4 (MNIST), 0.6 to 2.0 (Adult)
    The set of scaling factors in Algorithm 3.2; chosen manually as a reasonable range for each dataset.
  • q (repeated partial trainings) = 5
    Number of repeats in Algorithm 3.2 to average errors; the paper notes no significant change for larger values.
assumptions (4)
  • ad hoc to paper High condition number of a stacked weight matrix indicates the corresponding layer is over-proportioned, i.e., has too many neurons relative to other layers.
    This is the core heuristic underlying Algorithm 3.1, stated as an insight in Section 3 without proof.
  • domain assumption Condition numbers of partially trained networks (eta <= 3 epochs) are stable and predictive of the final network's structure.
    The method relies on partial training; Section 5.5 presents empirical evolution tables but explicitly says 'we cannot guarantee that one (or half) epoch will be adequate for all datasets'.
  • ad hoc to paper The scoring function 2*eps_v - eps_tr used in Algorithm 3.2 identifies the model with the best generalization among the scaled candidates.
    Used to select beta; it is a standard-looking bias-variance heuristic but is not derived and is evaluated only on the two datasets.
  • standard math Standard facts about SVD, QR, pivoted QR, and condition numbers of matrices.
    Used throughout for rank estimation and neuron selection; unproblematic.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Refining the Structure of Neural Networks Using Matrix Conditioning." pith.science (2026). https://pith.science/paper/3TIN7EKX

@misc{pith2026190802400,
  author       = {Pith},
  title        = {Pith review of: Refining the Structure of Neural Networks Using Matrix Conditioning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3TIN7EKX}},
  note         = {Machine review of arXiv:1908.02400}
}
read the original abstract

Deep learning models have proven to be exceptionally useful in performing many machine learning tasks. However, for each new dataset, choosing an effective size and structure of the model can be a time-consuming process of trial and error. While a small network with few neurons might not be able to capture the intricacies of a given task, having too many neurons can lead to overfitting and poor generalization. Here, we propose a practical method that employs matrix conditioning to automatically design the structure of layers of a feed-forward network, by first adjusting the proportion of neurons among the layers of a network and then scaling the size of network up or down. Results on sample image and non-image datasets demonstrate that our method results in small networks with high accuracies. Finally, guided by matrix conditioning, we provide a method to effectively squeeze models that are already trained. Our techniques reduce the human cost of designing deep learning models and can also reduce training time and the expense of using neural networks for applications.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 30 canonical work pages

  1. [1]

    J. M. Alvarez and M. Salzmann , Learning the number of neurons in deep networks , in Advances in Neural Information Processing Systems (NeurIPS 2016), 2016, pp. 2270–2278

    work page 2016

  2. [2]

    J. M. Alvarez and M. Salzmann , Compression-aware training of deep networks, in Advances in Neural Information Processing Systems (NeurIPS 2017), 2017, pp. 856–867

    work page 2017

  3. [3]

    Anders and O

    U. Anders and O. Korn , Model selection in neural networks , Neural Networks, 12 (1999), pp. 309–323

    work page 1999

  4. [4]

    Baker, O

    B. Baker, O. Gupta, N. Naik, and R. Raskar , Designing neural network architectures using reinforcement learning, arXiv preprint arXiv:1611.02167, (2016)

    arXiv 2016

  5. [5]

    Bender, P.-J

    G. Bender, P.-J. Kindermans, B. Zoph, V. V asudevan, and Q. Le , Understanding and simplifying one-shot architecture search, in International Conference on Machine Learning (ICML 2018), 2018, pp. 549–558

    work page 2018

  6. [6]

    Canziani, A

    A. Canziani, A. Paszke, and E. Culurciello , An analysis of deep neural network models for practical applications, arXiv preprint arXiv:1605.07678, (2016)

    arXiv 2016

  7. [7]

    Cashman, A

    D. Cashman, A. Perer, and H. Strobelt , Mast: A tool for visualizing CNN model ar- chitecture searches, in ICLR 2019 Debugging Machine Learning Models Workshop, 2019, https://debug-ml-iclr2019.github.io/cameraready/DebugML-19 paper 24.pdf

    work page 2019

  8. [8]

    T. F. Chan , Rank revealing QR factorizations, Linear Algebra and its Applications, 88 (1987), pp. 67–82

    work page 1987

Show all 38 references
  1. [9]

    Chung and T

    J. Chung and T. Shin , Simplifying deep neural networks for neuromorphic architectures , in 53nd ACM/EDAC/IEEE Design Automation Conference (DAC), 2016, pp. 1–6

    2016

  2. [10]

    E. L. Denton, W. Zaremba, J. Bruna, Y. LeCun, and R. Fergus , Exploiting linear structure REFINING THE STRUCTURE OF NEURAL NETWORKS 21 within convolutional networks for efficient evaluation , in Advances in Neural Information Processing Systems (NeurIPS 2014), 2014, pp. 1269–1277

    2014

  3. [11]

    Dua and C

    D. Dua and C. Graff , UCI machine learning repository, 2017, http://archive.ics.uci.edu/ml

    2017

  4. [12]

    S. A. Friedler, C. Scheidegger, S. Venkatasubramanian, S. Choudhary, E. P. Hamil- ton, and D. Roth , A comparative study of fairness-enhancing interventions in machine learning, in Proceedings of the Conference on Fairness, Accountability, and Transparency, FAT* ’19, 2019, pp. 329–338

    2019

  5. [13]

    Goetschalckx, P

    K. Goetschalckx, P. W ambacq, B. Moons, and M. Verhelst , Efficiently combining SVD, pruning, clustering and retraining for enhanced neural network compression , in Proceed- ings of the 2018 International Workshop on Embedded and Mobile Deep Learning, ACM, 2018, pp. 1–6

    2018

  6. [14]

    G. H. Golub and C. F. V an Loan , Matrix Computations , JHU Press, Baltimore, 4th ed., 2012

    2012

  7. [15]

    S. Han, J. Pool, J. Tran, and W. Dally , Learning both weights and connections for efficient neural network, in Advances in Neural Information Processing Systems (NeurIPS 2015), 2015, pp. 1135–1143

    2015

  8. [16]

    A. G. Howard, M. Zhu, B. Chen, D. Kalenichenko, W. W ang, T. Weyand, M. An- dreetto, and H. Adam , Mobilenets: Efficient convolutional neural networks for mobile vision applications, arXiv preprint arXiv:1704.04861, (2017)

    2017 arXiv

  9. [17]

    H. Hu, J. Langford, R. Caruana, S. Mukherjee, E. Horvitz, and D. Dey , Efficient forward architecture search, arXiv preprint arXiv:1905.13360, (2019)

    2019 arXiv

  10. [18]

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

    2016 arXiv

  11. [19]

    Jiang, N

    Y. Jiang, N. Bosch, R. S. Baker, L. Paquette, J. Ocumpaugh, J. M. A. L. Andres, A. L. Moore, and G. Biswas , Expert feature-engineering vs. Deep neural networks: Which is better for sensor-free affect detection?, in International Conference on Artificial Intelligence in Educatio...

    2018

  12. [20]

    C. Liu, B. Zoph, M. Neumann, J. Shlens, W. Hua, L.-J. Li, L. Fei-Fei, A. Yuille, J. Huang, and K. Murphy , Progressive neural architecture search , in European Con- ference on Computer Vision (ECCV), 2018, pp. 19–34

    2018

  13. [21]

    R. K. Mothilal, A. Sharma, and C. Tan , Explaining machine learning classifiers through diverse counterfactual explanations , arXiv preprint arXiv:1905.07697, (2019)

    2019 arXiv

  14. [22]

    Murray and D

    K. Murray and D. Chiang , Auto-sizing neural networks: With applications to n-gram lan- guage models, arXiv preprint arXiv:1508.05051, (2015)

    2015 arXiv

  15. [23]

    Neyshabur, Z

    B. Neyshabur, Z. Li, S. Bhojanapalli, Y. LeCun, and N. Srebro , The role of over- parametrization in generalization of neural networks, in International Conference on Learn- ing Representations (ICLR 2019), 2019

    2019

  16. [24]

    S. J. Nowlan and G. E. Hinton , Simplifying neural networks by soft weight-sharing , Neural Computation, 4 (1992), pp. 473–493

    1992

  17. [25]

    H. Pham, M. Guan, B. Zoph, Q. Le, and J. Dean , Efficient neural architecture search via parameters sharing, in International Conference on Machine Learning (ICML 2018), vol. 80, 2018, pp. 4095–4104

    2018

  18. [26]

    D. C. Psichogios and L. H. Ungar , SVD-NET: An algorithm that automatically selects network structure, IEEE Transactions on Neural Networks, 5 (1994), pp. 513–515

    1994

  19. [27]

    Sedghi, V

    H. Sedghi, V. Gupta, and P. M. Long , The singular values of convolutional layers , in Inter- national Conference on Learning Representations (ICLR 2019), 2019

    2019

  20. [28]

    Srivastava, G

    N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov , Dropout: A simple way to prevent neural networks from overfitting, The Journal of Machine Learning Research, 15 (2014), pp. 1929–1958

    2014

  21. [29]

    K. O. Stanley and R. Miikkulainen , Evolving neural networks through augmenting topolo- gies, Evolutionary Computation, 10 (2002), pp. 99–127. 22 R. YOUSEFZADEH AND D. P. O’LEARY

    2002

  22. [30]

    E. J. Teoh, K. C. Tan, and C. Xiang , Estimating the number of hidden neurons in a feed- forward network using the singular value decomposition , IEEE Transactions on Neural Networks, 17 (2006), pp. 1623–1629

    2006

  23. [31]

    Wistuba , Finding competitive network architectures within a day using UCT , arXiv pre- print arXiv:1712.07420, (2017)

    M. Wistuba , Finding competitive network architectures within a day using UCT , arXiv pre- print arXiv:1712.07420, (2017)

    2017 arXiv

  24. [32]

    Y. Xu, Y. Li, S. Zhang, W. Wen, B. W ang, Y. Qi, Y. Chen, W. Lin, and H. Xiong , Trained rank pruning for efficient deep neural networks , arXiv preprint arXiv:1812.02402, (2018)

    2018 arXiv

  25. [33]

    J. Xue, J. Li, and Y. Gong , Restructuring of deep neural network acoustic models with singular value decomposition, in Interspeech, 2013, pp. 2365–2369

    2013

  26. [34]

    Yousefzadeh and D

    R. Yousefzadeh and D. P. O’Leary , Debugging trained machine learning models using flip points , in ICLR 2019 Debugging Machine Learning Models Workshop, 2019, https: //debug-ml-iclr2019.github.io/cameraready/DebugML-19 paper 11.pdf

    2019

  27. [35]

    Zhang, S

    C. Zhang, S. Bengio, M. Hardt, B. Recht, and O. Vinyals , Understanding deep learning requires rethinking generalization, arXiv preprint arXiv:1611.03530, (2016)

    2016 arXiv

  28. [36]

    H. Zhou, J. M. Alvarez, and F. Porikli , Less is more: Towards compact CNNs, in European Conference on Computer Vision, Springer, 2016, pp. 662–677

    2016

  29. [37]

    Zoph and Q

    B. Zoph and Q. V. Le , Neural architecture search with reinforcement learning, arXiv preprint arXiv:1611.01578, (2016)

    2016 arXiv

  30. [38]

    B. Zoph, V. V asudevan, J. Shlens, and Q. V. Le , Learning transferable architectures for scal- able image recognition, in IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 8697–8710

    2018

Pith tools

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