Pith. sign in

REVIEW 3 major objections 4 minor 27 references

Building Efficient Lightweight CNN Models

T0 review · 3 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A two-branch training recipe produces a 14,862-parameter CNN that reaches 99% accuracy on MNIST.

desk verdict A small-CNN recipe worth a look, but the advertised 14,862-parameter count contradicts the architecture as described, and the 'state-of-the-art' label has no baseline to support it. read the letter →

arxiv 2501.15547 v1 pith:CVEWHFLI submitted 2025-01-26 cs.CV cs.AIcs.LG

Nathan Isong This is my paper · ORCID
classification cs.CVcs.AIcs.LG
keywords lightweightCNNdual-input-outputmodelprogressiveunfreezingtransferlearningMNISTfashionCIFAR-10compression
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 proposes a two-stage recipe for building very small convolutional networks from the ground up, rather than compressing a large pretrained model. First, two identical submodels are trained side by side, one on the original images and one on augmented copies, and their outputs are concatenated. Second, the trained weights are fused into a single model and fine-tuned by progressively unfreezing layers. The author reports that the resulting model contains only 14,862 parameters (0.17 MB) and reaches 99% accuracy on handwritten digit MNIST and 89% on fashion MNIST, suggesting that task-sized networks can match much larger models.

What carries the argument

The load-bearing mechanism is the dual-input-output construction: two identical feature extractors, one fed the original image and one fed a dynamically augmented version, are trained simultaneously and their outputs are concatenated. The dense layers are then converted to 1x1 convolutions so that the pre-trained weights survive the transition to a single-input model, and progressive unfreezing starting from the last layer fine-tunes the fused model. This combination is what the paper credits for the model's small size, robustness, and accuracy.

What would settle it

Train a single-branch CNN with the same 14,862 parameters and the same random rotation, zoom, brightness, and translation augmentation on MNIST and fashion MNIST. If it reaches 99% and 89% test accuracy, then the reported results do not depend on the dual-input-output design or progressive unfreezing, and the causal claim would be refuted.

Watch

Extended reading notes

Core claim

The central claim is that dual-input-output training followed by transfer learning with progressive unfreezing produces a unified single-input model that is both extremely small and accurate. The architecture uses two small convolutional submodels with 10 and 20 filters, concatenates their outputs, converts the dense layers into 1x1 convolutions to preserve learned weights, and adds a small fusion head before fine-tuning. The author reports test accuracies of 99% on handwritten digit MNIST, 89% on fashion MNIST, and 65% on CIFAR-10, with 14,862 and 19,622 trainable parameters respectively, and a model size of 0.17 MB (0.21 MB for CIFAR-10).

Load-bearing premise

The results are attributed to the specific dual-input plus progressive-unfreezing pipeline, but the paper does not compare against a plain small CNN trained with the same augmentation, so the method's unique contribution is not isolated.

Editorial extensions

If this is right

  • A model small enough to fit in 0.17 MB can reach 99% accuracy on MNIST, which is relevant for on-device and embedded inference.
  • The same pipeline can be applied to other small-image classification tasks by only changing the input shape, without structural modifications.
  • The method works without post-training quantization or pruning, so the reported size is the architecture's natural footprint.
  • On CIFAR-10 the accuracy drops to 65%, which indicates the method's limits on more complex, color-rich datasets.
  • Progressive unfreezing with a final SGD fine-tune is presented as the step that stabilizes the fused model and prevents overfitting.

Reading between the lines

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

  • If the causal claim holds, the paper implies a general recipe: instead of training one model on augmented data, train two branches on complementary views and then fuse them, which may be applicable to other domains such as audio or tabular data.
  • The 1x1 convolution conversion of dense layers suggests the method extends naturally to fully convolutional architectures and could be combined with semantic segmentation heads.
  • A direct head-to-head test against LeNet-5 with matched parameter counts would sharpen the 'state-of-the-art' claim, which the paper itself does not provide.
  • The CIFAR-10 result hints at a complexity ceiling: scaling to harder datasets would likely require more filters or depth, and a rule for that scaling is left open as future work.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes a two-stage training methodology for building small CNNs: first, a dual-input, dual-output model with two identical branches is trained, one branch on the original data and one on an augmented version; second, a single-input model is created by removing/replacing the branch dense layers, concatenating the remaining features, adding a new dense head, and fine-tuning with progressive unfreezing. The manuscript reports 99% test accuracy on MNIST, 89% on fashion MNIST, and 65% on CIFAR-10 with 14,862/19,622 trainable parameters and model sizes of about 0.17–0.21 MB, together with latency and throughput measurements. The abstract and introduction describe these results as state-of-the-art, although Section 4 explicitly states that the experiments are not directly compared with other models.

Significance. If fully supported, the paper would provide a useful data point: a roughly 15k-parameter model that reaches 99% on MNIST and 89% on fashion MNIST is a plausible and potentially useful efficiency result, and the reported test accuracies, classification reports, and confusion matrices are internally consistent. The held-out test-set evaluation is standard, and the paper is commendably transparent about several limitations, including the absence of direct comparisons. However, the central claims are not yet established: the described architecture in Section 3.5.2 does not match the parameter counts in Table 6, there is no baseline or ablation to support the causal attribution to the proposed method, and the 'state-of-the-art' label is unsupported. The contribution is primarily an empirical combination of existing techniques (augmentation, transfer learning, progressive unfreezing) rather than a new conceptual mechanism.

major comments (3)
  1. [§3.5.2 and Table 6] The reported parameter count of 14,862 does not match the architecture described in Section 3.5.2. For a 28×28×1 input, the branch feature map before the final dense layer is 5×5×20; retaining the original Dense(10) layers gives 2×(100+1,820+5,010)+672+330 = 14,862 parameters, whereas replacing those dense layers with 1×1 Conv2D(10) layers as stated gives 2×(100+1,820+210)+16,032+330 = 20,622 parameters. The CIFAR-10 count of 19,622 likewise matches the retained-dense model, not the 1×1-converted model. The paper therefore conflates two different models: the size claimed is for the original architecture, while the described 1×1 conversion produces a larger model. Please correct the text, the table, or provide a model artifact that resolves this discrepancy.
  2. [§4 and Abstract] The abstract's 'state-of-the-art accuracy' claim is not established by the body of the paper. Section 4 states that the results 'are not directly compared to other models,' and the only comparison offered is a qualitative remark about LeNet. Since 99% on MNIST is a routinely achieved accuracy and many models with comparable or smaller parameter counts exceed it, please replace the 'state-of-the-art' phrasing with a supported claim, or add a comparison table with relevant small CNN baselines.
  3. [§3.5 and §4] No ablation or baseline isolates the contribution of the dual-input design, the concatenation, or progressive unfreezing. For example, a single-branch CNN with the same parameter budget trained on augmented data might reach similar accuracy, which would reduce the methodological contribution to routine augmentation plus a small architecture. Please add at least a single-branch baseline trained on original data and a single-branch baseline trained on augmented data, and, if possible, a version of the proposed pipeline without the final progressive-unfreezing stage.
minor comments (4)
  1. [Throughout] The manuscript contains many typographical errors that should be corrected in a copyedit; examples include 'and and filter decomposition' in Section 1, 'T ransfer' and 'modther model' in several places, 'CIF AR-10' throughout, and 'rebuilded-model' in the final paragraph.
  2. [Figure 13 caption] The caption calls CIFAR-10 a 'handwritten' dataset; CIFAR-10 is not a handwritten dataset, so the caption should be corrected.
  3. [§3.5.3 Stage 1] The text says the model was trained for 20 epochs with early stopping, but the note in the Figure 9 caption says increasing epochs beyond 20 did not help; please clarify the final epoch count and the stopping criterion used for each dataset.
  4. [Table 1] Please state how 'Approx. Final Training Accuracy' is computed and whether it is the average over the six folds from Section 3.5.3, since the table currently gives no variance information.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: test-set accuracies are measured outputs, not derived from assumptions; the main concerns (uncompared SOTA claim, parameter-count inconsistency) are correctness issues, not circularity.

full rationale

All reported accuracies (99% MNIST, 89% Fashion MNIST, 65% CIFAR-10) come from held-out test-set evaluation with model selection on validation via ModelCheckpoint and early stopping; no result is computed from the method's definitions or from fitted constants. The architecture and two-stage training are described independently of the outcome. The 'state-of-the-art' adjective is unsupported because Section 4 explicitly says results 'are not directly compared to other models,' but this is an evidentiary weakness, not circular reasoning. The LeNet parameter comparison is a rough external benchmark, not a fitted input. No self-citation is load-bearing; references are standard external works (Howard and Ruder for progressive unfreezing, Pan and Yang for transfer learning, etc.). The reported parameter count of 14,862 appears inconsistent with the Section 3.5.2 1x1-conv replacement (a corrected count would be about 20,622), so the model whose size is claimed may not be the model whose accuracy is claimed; that is an internal-consistency or correctness issue, not a case of prediction reducing to inputs. The causal attribution of accuracy to the dual-input design is untested because no ablation exists, but absence of a control is not circularity. I find no step in which a derived quantity is equivalent by construction to its input.

Assumptions & free parameters 10 free parameters · 5 assumptions · 0 invented entities

The central claim rests on hand-chosen hyperparameters (10/20 filters, 32 hidden units, dropout 0.5, SGD settings, patience 5, six folds), unstated input wiring for the merged model, and standard supervised-learning assumptions. No formal derivation or ablation is provided, so these ledger entries carry the empirical burden.

free parameters (10)
  • Number of Conv2D filters in first block = 10
    Hand-chosen in Section 3.3; no ablation, but it fixes the parameter count and capacity of every reported model.
  • Number of Conv2D filters in second block = 20
    Hand-chosen as double the first block; no ablation.
  • Fusion dense layer units = 32
    Added in Section 3.5.2 before the softmax; no ablation.
  • Dropout rate in fusion head = 0.5
    Chosen to prevent overfitting; no sensitivity study.
  • SGD learning rate = 0.001
    Set in Section 3.5.3 stage 2; no learning-rate sweep.
  • SGD momentum = 0.9
    Chosen in Section 3.5.3; no sweep.
  • SGD weight decay = 1e-4
    Chosen in Section 3.5.3; no sweep.
  • Early stopping patience = 5
    Set in Section 3.4; affects training length and final weights.
  • Number of K-fold splits = 6
    Set in Section 3.5.3 stage 3; affects validation size and final model selection.
  • Augmentation ranges = rotation +-0.1 rad, zoom +-20%, translation +-20%, brightness factor 0
    Hand-picked in Section 3.2 with the caveat that no detailed technical consideration was applied; brightness factor 0 likely disables that augmentation.
assumptions (5)
  • domain assumption Standard supervised splits and labels of MNIST, Fashion-MNIST, and CIFAR-10 are reliable and representative.
    Section 3.1 uses the standard tf.keras dataset splits; no label-noise check is reported.
  • standard math Reshaping dense weights into 1x1 convolutional kernels preserves the learned representations.
    Invoked in Section 3.5.2 via Long et al.; if the reshape is not exactly reproducible the final model weights are under-specified.
  • domain assumption A single sparse categorical crossentropy loss applied to two outputs with default Keras weighting is an appropriate training objective.
    Section 3.4 specifies one loss and metric without output-specific weighting; this affects the dual-branch training dynamics.
  • standard math Early stopping and checkpoint selection on validation accuracy do not leak test information.
    Sections 3.4 and 3.5.3; test accuracies are reported after model selection, which is standard only if test data is never used for selection.
  • ad hoc to paper The merged two-branch model can be fine-tuned on original images alone without losing the dual-input benefit.
    Section 3.5.3 says further training used only the original dataset, but the final model still has two input branches; how both branches are fed in stages 2 and 3 is not stated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Building Efficient Lightweight CNN Models." pith.science (2026). https://pith.science/paper/CVEWHFLI

@misc{pith2026250115547,
  author       = {Pith},
  title        = {Pith review of: Building Efficient Lightweight CNN Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CVEWHFLI}},
  note         = {Machine review of arXiv:2501.15547}
}
read the original abstract

Convolutional Neural Networks (CNNs) are pivotal in image classification tasks due to their robust feature extraction capabilities. However, their high computational and memory requirements pose challenges for deployment in resource-constrained environments. This paper introduces a methodology to construct lightweight CNNs while maintaining competitive accuracy. The approach integrates two stages of training; dual-input-output model and transfer learning with progressive unfreezing. The dual-input-output model train on original and augmented datasets, enhancing robustness. Progressive unfreezing is applied to the unified model to optimize pre-learned features during fine-tuning, enabling faster convergence and improved model accuracy. The methodology was evaluated on three benchmark datasets; handwritten digit MNIST, fashion MNIST, and CIFAR-10. The proposed model achieved a state-of-the-art accuracy of 99% on the handwritten digit MNIST and 89% on fashion MNIST, with only 14,862 parameters and a model size of 0.17 MB. While performance on CIFAR-10 was comparatively lower (65% with less than 20,00 parameters), the results highlight the scalability of this method. The final model demonstrated fast inference times and low latency, making it suitable for real-time applications. Future directions include exploring advanced augmentation techniques, improving architectural scalability for complex datasets, and extending the methodology to tasks beyond classification. This research underscores the potential for creating efficient, scalable, and task-specific CNNs for diverse applications.

Figures

Figures reproduced from arXiv: 2501.15547 by the authors.

Figure 1
Figure 1. Flow Chart showing the methodology step. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Structure of model2 (identical to model1) before concatenation. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Concatenated model structure: here we have the inputs ( [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (16 more)
Figure 4
Figure 4. Figure 4: First training performance metrics of the two submodels for the handwritten [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: First training performance metrics of the two submodels for the fashion MNIST. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: First training performance metrics of the two submodels for the cifar10 dataset. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Final model architecture: mod1 f and mod2 f represent model1 and model2 without their dense layers; the next Conv2D layers on either side represent the converted dense layers. fine-tuning deep neural networks due to its ability to adapt weight adjustments with greater …
Figure 8
Figure 8. Figure 8: Second training performance metrics of the model for the MNIST handwritten [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Second training performance metrics of the model for the fashion MNIST dataset. [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Second training performance metrics of the model for the Cifar10 dataset. [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: Third training performance metrics of the two submodels for the MNIST hand [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]
Figure 12
Figure 12. Figure 12: Third training performance metrics of the two submodels for the fashion MNIST [PITH_FULL_IMAGE:figures/full_fig_p014_12.png]
Figure 13
Figure 13. Figure 13: Third training performance metrics of the two submodels for the cifar10 hand [PITH_FULL_IMAGE:figures/full_fig_p014_13.png]
Figure 14
Figure 14. Figure 14: Confusion matrix of the MNIST-trained model evaluated on the MNIST test [PITH_FULL_IMAGE:figures/full_fig_p016_14.png]
Figure 15
Figure 15. Figure 15: Confusion matrix of the fashion MNIST-trained model evaluated on the fashion [PITH_FULL_IMAGE:figures/full_fig_p017_15.png]
Figure 16
Figure 16. Figure 16: Confusion matrix of the cifar10-trained model evaluated on the Cifar10 test set. [PITH_FULL_IMAGE:figures/full_fig_p018_16.png]
Figure 17
Figure 17. Figure 17: MNIST CPU Throughput [PITH_FULL_IMAGE:figures/full_fig_p020_17.png]
Figure 20
Figure 20. Figure 20: fashion MNIST GPU Through￾put [PITH_FULL_IMAGE:figures/full_fig_p020_20.png]
Figure 22
Figure 22. Figure 22: CIFAR-10 GPU Throughput [PITH_FULL_IMAGE:figures/full_fig_p020_22.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 9 canonical work pages

  1. [1]

    Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, et al

    Martín Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, et al. Tensorflow: Large-scale machine learning on heterogeneous distributed systems. arXiv preprint arXiv:1603.04467, 2016. URL https://arxiv.org/abs/1603.04467. Accessed: 2025-01-14

    arXiv 2016

  2. [2]

    Practical recommendations for gradient-based training of deep architectures

    Yoshua Bengio. Practical recommendations for gradient-based training of deep architectures. arXiv preprint arXiv:1206.5533, 2012. URL https://arxiv.org/abs/1206.5533. Accessed: 2025-01-14

    arXiv 2012

  3. [3]

    Machine learning at microsoft with ml.net

    Eric Breck, Yi Ding, Vladimir Markov, Chris Sweeney, Yu Wei, and Pierre Bojanowski. Machine learning at microsoft with ml.net. arXiv preprint arXiv:1906.02243, 2020. URL https://arxiv.org/abs/1906.02243. Accessed: 2025-01-14

    arXiv 1906

  4. [4]

    Exploiting Linear Structure Within Convolutional Networks for Efficient Evaluation

    Emily Denton, Wojciech Zaremba, Joan Bruna, Yann LeCun, and Rob Fergus. Exploiting linear structure within convolutional networks for efficient evaluation. In Advances in Neural Information Processing Systems (NeurIPS), pages 1269--1277, 2014. URL https://arxiv.org/abs/1404.0736

    work page Pith review arXiv 2014

  5. [5]

    S. S. S. R. Depuru, B. A. Reddy, and S. K. Reddy. An experimental approach towards the performance assessment of optimizers in cnns. In 2019 International Conference on Computational Intelligence and Knowledge Economy (ICCIKE), pages 80--85. IEEE, 2019. doi:10.1109/ICCIKE47802.2019.9004342. URL https://ieeexplore.ieee.org/document/9004342

    arXiv 2019

  6. [6]

    Reducing model complexity in neural networks by using pyramid training approaches

    K van c S ah m Giray, Baha S en, Fatih Nar, and Ali \"O zg \"u n Ok. Reducing model complexity in neural networks by using pyramid training approaches. In Applied Sciences, volume 14, page 5898, 2024. doi:10.3390/app14135898. URL https://www.mdpi.com/2076-3417/14/13/5898

    work page doi:10.3390/app14135898 2024

  7. [7]

    Song Han, Jeff Pool, John Tran, and William J. Dally. Learning both weights and connections for efficient neural networks. In Advances in Neural Information Processing Systems (NeurIPS), pages 1135--1143, 2015. URL https://arxiv.org/abs/1506.02626

    arXiv 2015

  8. [8]

    Universal language model fine-tuning for text classification

    Jeremy Howard and Sebastian Ruder. Universal language model fine-tuning for text classification. arXiv preprint arXiv:1801.06146, 2018. URL https://arxiv.org/abs/1801.06146. Accessed: 2025-01-14

    arXiv 2018

Show all 27 references
  1. [9]

    Weinberger

    Gao Huang, Zhuang Liu, Laurens van der Maaten, and Kilian Q. Weinberger. Densely connected convolutional networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 2261--2269, 2018. doi:10.1109/CVPR.2017.243. URL https://arxiv.org/...

    2018 arXiv

  2. [10]

    Quantization and training of neural networks for efficient integer-arithmetic-only inference

    Benoit Jacob, Skirmantas Kligys, Bo Chen, Menglong Zhu, Matthew Tang, Andrew Howard, Hartwig Adam, and Dmitry Kalenichenko. Quantization and training of neural networks for efficient integer-arithmetic-only inference. In Proceedings of the IEEE Conference on Computer Vision an...

    2018

  3. [11]

    Efficient gpu usage tips

    Kaggle . Efficient gpu usage tips. https://www.kaggle.com/docs/efficient-gpu-usage, 2025. Accessed: 2025-01-14

    2025

  4. [12]

    A study of cross-validation and bootstrap for accuracy estimation and model selection

    Ron Kohavi. A study of cross-validation and bootstrap for accuracy estimation and model selection. In Proceedings of the 14th International Joint Conference on Artificial Intelligence (IJCAI), pages 1137--1145, 1995. URL https://www.ijcai.org/Proceedings/95-2/Papers/016.pdf

    1995

  5. [13]

    Imagenet classification with deep convolutional neural networks

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. In Advances in Neural Information Processing Systems, volume 25, pages 1097--1105. Curran Associates, Inc., 2012. doi:10.1145/3065386

    2012 doi

  6. [14]

    How to fine-tune vision models with sgd

    Ananya Kumar, Ruoqi Shen, Sebastien Bubeck, and Suriya Gunasekar. How to fine-tune vision models with sgd. arXiv preprint arXiv:2211.09359, 2022. URL https://arxiv.org/abs/2211.09359

    2022 arXiv

  7. [15]

    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. doi:10.1109/5.726791

    1998 doi

  8. [16]

    Retraining-based iterative weight quantization for deep neural networks

    Dongsoo Lee and Byeongwook Kim. Retraining-based iterative weight quantization for deep neural networks. arXiv preprint arXiv:1805.11233, 2018. URL https://arxiv.org/abs/1805.11233. Accessed: 2025-01-14

    2018 arXiv

  9. [17]

    Lightweight deep learning for resource-constrained environments

    Hou-I Liu et al. Lightweight deep learning for resource-constrained environments. arXiv preprint arXiv:2404.07236, 2022. URL https://arxiv.org/abs/2404.07236. Accessed: 2025-01-10

    2022 arXiv

  10. [18]

    Fully convolutional networks for semantic segmentation

    Jonathan Long, Evan Shelhamer, and Trevor Darrell. Fully convolutional networks for semantic segmentation. arXiv preprint arXiv:1411.4038, 2015. URL https://arxiv.org/abs/1411.4038. Accessed: 2025-01-14

    2015 arXiv

  11. [19]

    Maji and R

    S. Maji and R. Mullins. On the reduction of computational complexity of deep convolutional neural networks. In Entropy, volume 20, page 305, 2018. doi:10.3390/e20040305. URL https://www.mdpi.com/1099-4300/20/4/305

    2018 doi

  12. [20]

    A survey on transfer learning

    Sinno Jialin Pan and Qiang Yang. A survey on transfer learning. IEEE Transactions on Knowledge and Data Engineering, 22 0 (10): 0 1345--1359, 2010. doi:10.1109/TKDE.2009.191

    2010 doi

  13. [21]

    Khoshgoftaar

    Connor Shorten and Taghi M. Khoshgoftaar. A survey on image data augmentation for deep learning. Journal of Big Data, 6 0 (1): 0 60, 2019. doi:10.1186/s40537-019-0197-0. URL https://journalofbigdata.springeropen.com/articles/10.1186/s40537-019-0197-0

    2019 doi

  14. [22]

    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. doi:10.48550/arXiv.1409.1556

    2014 doi arXiv

  15. [23]

    Dropout: A simple way to prevent neural networks from overfitting

    Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 15 0 (1): 0 1929--1958, 2014. URL https://jmlr.org/papers/v15/srivastava14a.html

    1929

  16. [24]

    Rethinking the inception architecture for computer vision

    Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking the inception architecture for computer vision. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 2818--2826, 2015. doi:10.1109/CVPR.2015...

    2015

  17. [25]

    Gpu kernel-tesla p100 gpu

    Tee Yee. Gpu kernel-tesla p100 gpu. https://www.kaggle.com/code/teeyee314/gpu-kernel-tesla-p100-gpu, 2025. Accessed: 2025-01-14

    2025

  18. [26]

    Lcrm: Layer-wise complexity reduction method for cnn model optimization on end devices

    Wei Zhang, Ming Li, Xiaowei Chen, and Yanzhi Wang. Lcrm: Layer-wise complexity reduction method for cnn model optimization on end devices. In IEEE Transactions on Neural Networks and Learning Systems, volume 34, pages 3456--3467, 2023. doi:10.1109/TNNLS.2023.10168115. URL http...

    2023

  19. [27]

    The correct way to measure inference time of deep neural networks

    Chen Zhao. The correct way to measure inference time of deep neural networks. https://towardsdatascience.com/the-correct-way-to-measure-inference-time-of-deep-neural-networks-304a54e5187f, 2020. Accessed: 2025-01-06

    2020

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.