Pith. sign in

REVIEW 4 major objections 3 minor 48 references

Stress-Aware Resilient Neural Training

T0 review · 4 major / 3 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A stress-aware optimizer claims to escape sharp minima by injecting noise only when training stagnates, improving generalization at minimal cost.

desk verdict This submission's abstract and body are two different papers; the claimed ML method appears nowhere in the text, so there is nothing to referee. read the letter →

arxiv 2508.00098 v1 pith:62P4UARO submitted 2025-07-31 cs.LG cs.AIcs.CV

Ashkan Shakarami , Yousef Yeganeh , Azade Farshad , Lorenzo Nicole , Stefano Ghidoni , Nassir Navab This is my paper
classification cs.LGcs.AIcs.CV
keywords stress-awarelearningplasticdeformationoptimizeradaptivenoiseinjectionsharpminimaflatgeneralizationresilientneuraltraining
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

Stress-Aware Resilient Neural Training argues that an optimizer can be made resilient by borrowing a distinction from materials science: elastic (temporary) versus plastic (permanent) deformation. Its proposed Plastic Deformation Optimizer monitors an internal stress signal—defined as stagnation in training loss and accuracy—and injects adaptive noise into model parameters when that signal shows persistent optimization difficulty. The paper claims this lets a network escape sharp minima, converge to flatter and more generalizable regions, and improve robustness and generalization across six architectures, four optimizers, and seven vision benchmarks at minimal computational cost. The submitted full text, however, is an unrelated mathematics manuscript and does not contain the method, algorithm, or experiments described in the abstract, so this central claim cannot currently be checked from the submission.

What carries the argument

The load-bearing object is the Plastic Deformation Optimizer, a noise-injection mechanism triggered by an internal stress signal. The stress signal is defined as stagnation in training loss and accuracy, indicating persistent optimization difficulty; when it fires, adaptive noise is added to model parameters. This timing rule is the whole engine of the claimed effect: the stress signal decides when noise is injected, and the paper attributes the escape from sharp minima to that decision rather than to a fixed or constant noise schedule. The submitted full text provides none of this optimizer's definition or code, so the machinery exists only as described in the abstract.

What would settle it

Run the same architectures and benchmarks under three schedules with matched total noise: the stress-triggered schedule, a fixed-interval schedule, and a random-timing schedule. If either the fixed or random schedule matches the stress-triggered schedule in validation accuracy and in the flatness of the final minimum, the claimed causal role of the stress signal is refuted. Independently, an examination of the actual submission for the optimizer's code and experiments would show whether the reported seven-benchmark results exist at all.

Watch

Extended reading notes

Core claim

The central claim, stated in the paper's own terms, is that stagnation of training loss and accuracy can be read as a stress signal for persistent optimization difficulty. When that signal appears, the Plastic Deformation Optimizer reacts by injecting adaptive noise into the model parameters, which allows the model to escape sharp minima and move toward flatter regions of the loss landscape that generalize better. The authors present Stress-Aware Learning as a resilient training paradigm, with the elastic/plastic deformation analogy standing in for the difference between recoverable and permanent training states. They assert that this yields improved robustness and generalization on six architectures, four optimizers, and seven vision benchmarks with minimal computational overhead. The submitted full text does not match this narrative: it is an unrelated mathematical manuscript, so the abstract is the only available statement of the discovery.

Load-bearing premise

The load-bearing premise is that stagnation in training loss and accuracy reliably signals that the optimizer is stuck in a sharp minimum, so injecting noise at exactly that moment—rather than at any other moment—is what produces flatter, more generalizable solutions.

Editorial extensions

If this is right

  • Training runs that plateau on loss or accuracy could be rescued automatically, without a human choosing when to anneal or restart.
  • The final model would sit in flatter regions of the loss landscape, which typically transfer better to unseen data.
  • The mechanism would be portable: it should work with any of the four optimizers and six architectures evaluated, since it only watches the metrics, not the model's internal structure.
  • Computational overhead stays minimal, because noise injection is event-driven and only activates during stagnation rather than every step.

Reading between the lines

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

  • The paper does not establish that the stress trigger is the cause of the gains; the same benefits might come from injecting comparable noise at fixed intervals or at random times. An ablation comparing matched total noise across trigger timing would isolate this.
  • If trigger timing turns out to be irrelevant, the elastic/plastic stress framing reduces to existing parameter-noise or label-noise methods, and the contribution would be an engineering variant rather than a new training paradigm.
  • Because the submitted full text is a different manuscript entirely, the experimental numbers cited in the abstract currently stand on the abstract alone; recovering the actual implementation would be the first step in verifying or falsifying the claim.
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

4 major / 3 minor

Summary. The abstract describes Stress-Aware Learning, a resilience-oriented neural training paradigm, and a Plastic Deformation Optimizer that injects adaptive noise into model parameters when an internal stress signal indicates stagnation in training loss and accuracy, with claimed improvements in robustness and generalization across six architectures, four optimizers, and seven vision benchmarks. The full text, however, is the mathematics paper 'Holonomy of affine surfaces' by Paul Apisa, Matt Bainbridge, and Jane Wang, with a running header referencing arXiv:2508.00100. The body contains no description of the Plastic Deformation Optimizer, no stress signal definition, no noise schedule, no training protocol, no benchmark tables, and no experimental results. The submitted document therefore does not contain the work promised by the abstract, and the central claim cannot be verified or falsified from the manuscript as submitted.

Significance. If the abstract's claims were supported, the contribution would be relevant to robust and generalizable neural network training: a stress-triggered noise injection mechanism for escaping sharp minima is a plausible and testable idea. However, the submission provides no evidence for any of these claims. There are no machine-checked proofs, no reproducible code, no parameter-free derivations, and no experimental tables for the described method. The mathematical content of the body, while presumably of interest in its own field, does not bear on the machine-learning claim. The only falsifiable prediction in the abstract, the benchmark improvement, is unverifiable because the experiments are not reported. The manuscript as submitted is therefore not a reviewable version of the claimed work.

major comments (4)
  1. [Abstract vs. full text] The full text is not the Stress-Aware Learning paper advertised in the abstract. It is the mathematics paper 'Holonomy of affine surfaces' by Apisa, Bainbridge, and Wang, as is evident from the title, the author line, the section headings, and the running header 'arXiv:2508.00100'. The body contains no mention of the Plastic Deformation Optimizer, stress signals, adaptive noise, sharp minima, training loss, or benchmarks. This mismatch removes the entire evidential basis for the abstract's central claim, so the central claim cannot be verified or falsified from the submitted document.
  2. [Abstract (experiments claim)] The abstract claims improved robustness and generalization across six architectures, four optimizers, and seven vision benchmarks, but no experimental protocol, no tables, no error bars, and no description of the optimizer appear anywhere in the body. The headline performance claim is therefore entirely ungrounded in the submitted text.
  3. [Abstract (stress signal definition)] The abstract defines the stress signal by stagnation in training loss and accuracy, which are the same quantities the method is intended to improve. This creates a design-level circularity: the trigger and the target are the same metric. Without ablations that vary only the timing of noise injection, such as stress-triggered versus fixed-schedule or random-timing injection with matched total noise, the causal attribution of the reported gains to the stress signal is not established. The submission contains no such ablations.
  4. [Abstract (free parameters)] The method's load-bearing free parameters, namely the stress threshold on loss and accuracy stagnation, the adaptive noise scale, and the stress observation window or evaluation frequency, are not defined anywhere in the submission. Consequently, the claims of 'adaptive noise' and 'minimal computational overhead' cannot be quantitatively assessed.
minor comments (3)
  1. [Abstract] The GitHub URL for the code and 3D visuals cannot be checked because no code or supplementary material is included in the submission.
  2. [Full text (running header)] The running header 'arXiv:2508.00100' and the title 'Holonomy of affine surfaces' are inconsistent with the submission's abstract and arXiv identifier, which further confirms that the body text belongs to a different document.
  3. [References] The reference list contains no entries related to neural network training, loss landscape geometry, sharp or flat minima, or stochastic optimization, which is additional evidence that the full text is not the described paper.

Circularity Check

0 steps flagged · score 0.0 of 10

No derivational circularity found; the abstract's ML claim is unverifiable because the submitted full text is a different mathematics paper.

full rationale

The full text supplied with arXiv:2508.00098 is the pure-mathematics paper 'Holonomy of affine surfaces' (arXiv:2508.00100), not the Stress-Aware Learning paper described in the abstract. I therefore cannot walk a derivation chain for the Plastic Deformation Optimizer: there is no stress signal, no noise schedule, no sharp-minima argument, and no experimental protocol in the body text. Absence of evidence is not circularity. The mathematical paper itself is not circular: its theorems (1.1-1.6) are proved from explicit hypercohomology and short-exact-sequence calculations, and its self-citations to [ABW23] are used as published definitions and background rather than as the source of the new results. The only self-referential element is the abstract's definition of 'stress signal' as stagnation in training loss and accuracy, which are also the metrics one would use to measure improvement; but this is a heuristic coupling in an adaptive schedule, not a formal reduction in which the predicted quantity equals the fitted input. Since the method is absent, the central claim is unverifiable rather than circular, so the appropriate circularity score is 0.

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

The ledger is assembled from the abstract only, because the full text is a mismatched mathematics paper and contains none of the proposed method. The central claim depends on at least three unspecified hyperparameters (stress threshold, noise scale, observation window), on the assumption that training stagnation diagnoses sharp-minimum entrapment, and on the contested premise that flatter minima generalize better. The stress signal and the elastic/plastic framing are internal constructs without independent falsifiable handles. The math text that actually appears relies on the authors' prior paper [ABW23] for the definition of the moduli space and concedes that its Theorem 1.5 is implied by Serandour's work.

free parameters (3)
  • stress threshold on loss/accuracy stagnation
    The abstract does not state how much stagnation in training loss and accuracy is required to trigger the noise injection; this threshold is a hand-chosen or tuned hyperparameter on which the behavior of the optimizer depends.
  • adaptive noise scale
    The abstract says noise is 'adaptive' but reports neither the base noise magnitude nor how it adapts; the scale of the injected parameter noise is a fitted or hand-chosen quantity that determines whether the method escapes minima or destroys learned weights.
  • stress observation window and evaluation frequency
    How often the stress signal is evaluated (per step, per epoch) and over what lookback window is unspecified; this timing choice materially affects when noise is injected.
assumptions (3)
  • domain assumption Stagnation in training loss and accuracy is a reliable proxy for persistent optimization difficulty caused by sharp minima.
    Invoked in the abstract's definition of the internal stress signal; no evidence is provided that stagnation correlates with sharp-minimum entrapment rather than with learning-rate schedules, data noise, or capacity limits.
  • domain assumption Escaping sharp minima and converging to flatter regions improves generalization.
    The abstract asserts the mechanism ('escape sharp minima and converge toward flatter, more generalizable regions'); the flatness-generalization link is a contested empirical hypothesis in the literature and is not demonstrated in the provided text.
  • domain assumption Injecting isotropic parameter noise preserves the useful structure of the trained model while perturbing it enough to escape minima.
    The abstract assumes the noise perturbation does not destroy learned features; no analysis of the perturbation's effect on the loss landscape is present.
invented entities (2)
  • Stress signal (training stress)
    purpose: Internal scalar computed from stagnation in training loss and accuracy; triggers adaptive noise injection.
    The signal is defined entirely in terms of the training run's own metrics and has no falsifiable handle outside the paper; the abstract does not specify its functional form.
  • Temporary (Elastic) and Permanent (Plastic) Deformation states
    purpose: Conceptual framing borrowed from materials science to label stable versus stressed training regimes.
    A metaphorical construct; no operational definition or measurable distinction between elastic and plastic regimes is provided in the abstract.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Stress-Aware Resilient Neural Training." pith.science (2026). https://pith.science/paper/62P4UARO

@misc{pith2026250800098,
  author       = {Pith},
  title        = {Pith review of: Stress-Aware Resilient Neural Training},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/62P4UARO}},
  note         = {Machine review of arXiv:2508.00098}
}
read the original abstract

This paper introduces Stress-Aware Learning, a resilient neural training paradigm in which deep neural networks dynamically adjust their optimization behavior - whether under stable training regimes or in settings with uncertain dynamics - based on the concept of Temporary (Elastic) and Permanent (Plastic) Deformation, inspired by structural fatigue in materials science. To instantiate this concept, we propose Plastic Deformation Optimizer, a stress-aware mechanism that injects adaptive noise into model parameters whenever an internal stress signal - reflecting stagnation in training loss and accuracy - indicates persistent optimization difficulty. This enables the model to escape sharp minima and converge toward flatter, more generalizable regions of the loss landscape. Experiments across six architectures, four optimizers, and seven vision benchmarks demonstrate improved robustness and generalization with minimal computational overhead. The code and 3D visuals will be available on GitHub: https://github.com/Stress-Aware-Learning/SAL.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

48 extracted references · 45 canonical work pages

  1. [1]

    An overview of gradient descent optimization algorithms, 2016

    Sebastian Ruder. An overview of gradient descent optimization algorithms, 2016

    work page 2016

  2. [2]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 , 2014

    arXiv 2014

  3. [3]

    An overview of gradient descent optimization algorithms

    Sebastian Ruder. An overview of gradient descent optimization algorithms. arXiv preprint arXiv:1609.04747 , 2016

    arXiv 2016

  4. [4]

    Incorporating nesterov momentum into adam

    Timothy Dozat. Incorporating nesterov momentum into adam. ICLR Workshop, 2016

    work page 2016

  5. [5]

    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(56):1929--1958, 2014

    1929

  6. [6]

    Deep networks with stochastic depth

    Gao Huang, Yu Sun, Zhuang Liu, Daniel Sedra, and Kilian Q Weinberger. Deep networks with stochastic depth. In European Conference on Computer Vision (ECCV) , pages 646--661. Springer, 2016

    work page 2016

  7. [7]

    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 CVPR , pages 2818--2826, 2016

    work page 2016

  8. [8]

    Sharpness-aware minimization for efficiently improving generalization

    Pierre Foret, Ariel Kleiner, Hossein Mobahi, and Behnam Neyshabur. Sharpness-aware minimization for efficiently improving generalization. In ICLR , 2020

    work page 2020

Show all 48 references
  1. [9]

    Averaging weights leads to wider optima and better generalization

    Pavel Izmailov, Dmitry Podoprikhin, Timur Garipov, Dmitry Vetrov, and Andrew Gordon Wilson. Averaging weights leads to wider optima and better generalization. UAI , 2018

    2018

  2. [10]

    Entropy-sgd: Biasing gradient descent into wide valleys

    Pratik Chaudhari, Anna Choromanska, Stefano Soatto, Yann LeCun, Carlo Baldassi, Christian Borgs, Jennifer Chayes, Levent Sagun, and Riccardo Zecchina. Entropy-sgd: Biasing gradient descent into wide valleys. In International Conference on Learning Representations (ICLR) , 2017

    2017

  3. [11]

    Bayesian learning via stochastic gradient langevin dynamics

    Max Welling and Yee Whye Teh. Bayesian learning via stochastic gradient langevin dynamics. Proceedings of the 28th International Conference on Machine Learning (ICML) , pages 681--688, 2011

    2011

  4. [12]

    Noise can help: Accelerating learning through controlled noise injection

    Qizhe Xie, Zihang Dai, Eduard Hovy, Minh-Thang Luong, and Quoc V Le. Noise can help: Accelerating learning through controlled noise injection. In ICML , 2020

    2020

  5. [13]

    Training robust deep networks with adversarial noise

    Yuandong Liu et al. Training robust deep networks with adversarial noise. In NeurIPS , 2021

    2021

  6. [14]

    Online learning rate adaptation with hypergradient descent

    Atilim Gunes Baydin, Robert Cornish, Mark Rubio, David Martínez-Rubio, and Barak A Pearlmutter. Online learning rate adaptation with hypergradient descent. ICLR , 2018

    2018

  7. [15]

    Model-agnostic meta-learning for fast adaptation of deep networks

    Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation of deep networks. In ICML , 2017

    2017

  8. [16]

    Towards deep learning models resistant to adversarial attacks

    Aleksander Madry, Aleksandar Makelov, Ludwig Schmidt, Dimitris Tsipras, and Adrian Vladu. Towards deep learning models resistant to adversarial attacks. In ICLR , 2018

    2018

  9. [17]

    Certifiable distributional robustness with principled adversarial training

    Aman Sinha, Hongseok Namkoong, and John Duchi. Certifiable distributional robustness with principled adversarial training. In ICLR , 2018

    2018

  10. [18]

    Improving dnn robustness to adversarial attacks using jacobian regularization

    Daniel Jakubovitz and Raja Giryes. Improving dnn robustness to adversarial attacks using jacobian regularization. In ECCV , 2018

    2018

  11. [19]

    N. Nie, L. Su, G. Deng, H. Li, H. Yu, and A. K. Tieu. A review on plastic deformation induced surface/interface roughening of sheet metallic materials. Journal of Materials Research and Technology , 15:6574--6607, 2021

    2021

  12. [20]

    K. B. Kim, T. H. Kim, and E. H. Lee. Effect of formulation method for plastic deformation rate on topology optimization considering elastic-plastic behavior. European Journal of Mechanics-A/Solids , 106:105347, 2024

    2024

  13. [21]

    Y. H. Li, F. Shen, M. A. Güler, and L. L. Ke. Modeling multi-physics electrical contact on rough surfaces considering elastic-plastic deformation. International Journal of Mechanical Sciences , 269:109066, 2024

    2024

  14. [22]

    J. Tang, G. Lei, Q. Wu, L. Zhang, and F. Ning. An improved analytical model of effective thermal conductivity for hydrate-bearing sediments during elastic-plastic deformation and local thermal stimulation. Energy , 305:132293, 2024

    2024

  15. [23]

    Particle swarm optimization

    James Kennedy and Russell Eberhart. Particle swarm optimization. In Proceedings of ICNN'95 - International Conference on Neural Networks , volume 4, pages 1942--1948. IEEE, 1995

    1942

  16. [24]

    John H. Holland. Adaptation in Natural and Artificial Systems . University of Michigan Press, 1975

    1975

  17. [25]

    Grey wolf optimizer

    Seyedali Mirjalili, Seyed Mohammad Mirjalili, and Andrew Lewis. Grey wolf optimizer. Advances in Engineering Software , 69:46--61, 2014

    2014

  18. [26]

    Daniel Gelatt, and Mario P

    Scott Kirkpatrick, C. Daniel Gelatt, and Mario P. Vecchi. Optimization by simulated annealing. Science , 220(4598):671--680, 1983

    1983

  19. [27]

    Depvit-cad: Deployable vision transformer-based cancer diagnosis in histopathology

    Ashkan Shakarami, Lorenzo Nicol \`e , Rocco Cappellesso, Angelo Paolo Dei Tos, and Stefano Ghidoni. Depvit-cad: Deployable vision transformer-based cancer diagnosis in histopathology. arXiv preprint arXiv:2507.10250 , 2025

    2025 arXiv

  20. [28]

    Unit-based histopathology tissue segmentation via multi-level feature representation

    Ashkan Shakarami, Azade Farshad, Yousef Yeganeh, Lorenzo Nicol \`e , Patrick Sch \"u ffler, Stefano Ghidoni, and Nassir Navab. Unit-based histopathology tissue segmentation via multi-level feature representation. arXiv preprint arXiv:2507.12427 , 2025

    2025

  21. [29]

    Transformative ai for automating histopathology workflows, 2025

    Ashkan Shakarami. Transformative ai for automating histopathology workflows, 2025. Manuscript in preparation

    2025

  22. [30]

    Velu: Variance-enhanced learning unit for deep neural networks

    Ashkan Shakarami, Yousef Yeganeh, Azade Farshad, Lorenzo Nicol \`e , Stefano Ghidoni, and Nassir Navab. Velu: Variance-enhanced learning unit for deep neural networks. arXiv preprint arXiv:2504.15051 , 2025

    2025

  23. [31]

    Ai for advanced cancer diagnosis: a cad system empowered by a novel vision transformer network for histopathology analysis

    Ashkan Shakarami, Lorenzo Nicole, and Stefano Ghidoni. Ai for advanced cancer diagnosis: a cad system empowered by a novel vision transformer network for histopathology analysis. Virchows Archiv , 485:S397--S398, 2024

    2024

  24. [32]

    A cad system for diagnosing alzheimer’s disease using 2d slices and an improved alexnet-svm method

    Ashkan Shakarami, Hamed Tarrah, and Amir Mahdavi-Hormat. A cad system for diagnosing alzheimer’s disease using 2d slices and an improved alexnet-svm method. Optik , 212:164237, 2020

    2020

  25. [33]

    Tcnn: A transformer convolutional neural network for artifact classification in whole slide images

    Ashkan Shakarami, Lorenzo Nicole, Marco Terreran, Angelo Paolo Dei Tos, and Stefano Ghidoni. Tcnn: A transformer convolutional neural network for artifact classification in whole slide images. Biomedical Signal Processing and Control , 84:104812, 2023

    2023

  26. [34]

    A fast and yet efficient yolov3 for blood cell detection

    Ashkan Shakarami, Mohammad Bagher Menhaj, Amir Mahdavi-Hormat, and Hamed Tarrah. A fast and yet efficient yolov3 for blood cell detection. Biomedical Signal Processing and Control , 66:102495, 2021

    2021

  27. [35]

    Densely connected convolutional networks

    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 4700--4708, 2017

    2017

  28. [36]

    Mobilenetv2: Inverted residuals and linear bottlenecks

    Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen. Mobilenetv2: Inverted residuals and linear bottlenecks. In Proceedings of the IEEE conference on computer vision and pattern recognition (CVPR) , pages 4510--4520, 2018

    2018

  29. [37]

    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 (CVPR) , pages 770--778, 2016

    2016

  30. [38]

    Efficientnet: Rethinking model scaling for convolutional neural networks

    Mingxing Tan and Quoc Le. Efficientnet: Rethinking model scaling for convolutional neural networks. In International conference on machine learning (ICML) , pages 6105--6114. PMLR, 2019

    2019

  31. [39]

    Imagenette: A smaller subset of imagenet for fast experimentation

    Jeremy Howard. Imagenette: A smaller subset of imagenet for fast experimentation. https://github.com/fastai/imagenette, 2019

    2019

  32. [40]

    Imagewoof: A variant of imagenette with similar classes

    Jeremy Howard. Imagewoof: A variant of imagenette with similar classes. https://github.com/fastai/imagenette, 2019

    2019

  33. [41]

    Tiny imagenet challenge

    Ya Le. Tiny imagenet challenge. In CS231N Course Project Report , 2015. https://tiny-imagenet.herokuapp.com/

    2015

  34. [42]

    Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification

    Patrick Helber, Benjamin Bischke, Andreas Dengel, and Damian Borth. Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 12(7):2217--2226, 2019

    2019

  35. [43]

    Deeper, broader and artier domain generalization

    Da Li, Yongxin Yang, Yi-Zhe Song, and Timothy M Hospedales. Deeper, broader and artier domain generalization. In Proceedings of the IEEE International Conference on Computer Vision (ICCV) , pages 5542--5550, 2017

    2017

  36. [44]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky and Geoffrey Hinton. Learning multiple layers of features from tiny images. In Technical Report, University of Toronto , 2009

    2009

  37. [45]

    Y. Wang, X. Li, Y. Wu, and M. Wu. An improved image feature representation method for image classification. Multimedia Tools and Applications , 77(15):19515--19534, 2018

    2018

  38. [46]

    Lecture 6.5—rmsprop: Divide the gradient by a running average of its recent magnitude, 2012

    Tijmen Tieleman and Geoffrey Hinton. Lecture 6.5—rmsprop: Divide the gradient by a running average of its recent magnitude, 2012. COURSERA: Neural Networks for Machine Learning

    2012

  39. [47]

    An effective reinforcement learning method for preventing the overfitting of convolutional neural networks

    Amir Mahdavi-Hormat, Mohammad Bagher Menhaj, and Ashkan Shakarami. An effective reinforcement learning method for preventing the overfitting of convolutional neural networks. Advances in Computational Intelligence , 2(5):34, 2022

    2022

  40. [48]

    Scenegenie: Scene graph guided diffusion models for image synthesis

    Azade Farshad, Yousef Yeganeh, Yuwei Chi, Chunhua Shen, Bj \"o rn Ommer, and Nassir Navab. Scenegenie: Scene graph guided diffusion models for image synthesis. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pages 88--98, 2023

    2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.