Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

Thermal Image-based Fault Diagnosis in Induction Machines via Self-Organized Operational Neural Networks

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Shallow 3-layer Self-ONN beats EfficientNet on thermal motor fault diagnosis

desk verdict Plausible efficiency result for 2D Self-ONNs on thermal motor images, but the headline accuracy claim over EfficientNet B1 is not supported by the current experimental design. read the letter →

arxiv 2412.05901 v1 pith:E27M6KAM submitted 2024-12-08 cs.LG cs.CVeess.IV

Sertac Kilickaya , Cansu Celebioglu , Levent Eren , Murat Askar This is my paper · ORCID
classification cs.LGcs.CVeess.IV
keywords thermalimagingfaultdiagnosisinductionmachinesSelf-OrganizedOperationalNeuralNetworksconvolutionalbrokenrotormisalignmentedgecomputing
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 sets out to show that a very shallow 2D Self-Organized Operational Neural Network can diagnose induction-motor faults from thermal images at least as accurately as much deeper CNN models. Its best configuration, a three-operational-layer Self-ONN with Taylor-polynomial order Q=2, reaches 0.942 ± 0.070 accuracy on the healthy/misalignment/broken-rotor classification task, marginally above the best CNN baseline, EfficientNet B1 at 0.932 ± 0.086. That accuracy comes with only 294,083 trainable parameters and a 0.61 ms average inference time, versus 6,517,027 parameters and 14.97 ms for EfficientNet B1. If the comparison holds, the practical payoff is a non-contact condition-monitoring system that runs in near real time on edge hardware without sacrificing diagnostic accuracy.

What carries the argument

The load-bearing mechanism is the self-organized operational (generative) neuron, which computes pre-activations as $f_{\psi_k^l}(w_k^{(Q)}(r,t), y^{l-1}(m+r,n+t)) = \sum_{q=1}^Q w_k^{(Q)}(r,t,q)\,(y^{l-1})^{q}$, realized as Q ordinary 2D convolutions applied to powers of the layer input and summed. The order Q is a hyperparameter controlling the polynomial degree; Q=1 reduces the layer to a standard convolution, while Q>1 injects non-linearity inside the neuron, and all weights are learned by standard backpropagation. This mechanism is what lets the three-layer network express richer mappings than a depth-matched CNN.

What would settle it

Run the same 5-fold protocol with a single preprocessing pipeline and optimizer for both a plain 2D CNN (Q=1) and a Self-ONN (Q=2) matched in parameter count, and check whether the accuracy gap observed here (0.942 vs 0.887) persists; if it shrinks to less than one standard deviation, the attributed non-linearity benefit is not the cause.

Watch

Extended reading notes

Core claim

The central claim is that the generative neuron, whose nodal operator is a Taylor-polynomial approximation of degree Q over the input patch, can substitute for the depth of conventional CNNs in this diagnostic task. With Q=2, eight filters per layer, and kernel sizes 5×5, 3×3, and 2×2, the network reaches the best mean accuracy, F1, and recall among all compared models, and its confusion matrix shows zero errors between misalignment and broken rotor classes. The same network forced to Q=1, which is mathematically a plain 2D CNN, drops to 0.887 ± 0.106 accuracy, so the paper attributes the gain to the non-linear operational neurons rather than to the particular filter sizes or pooling scheme.

Load-bearing premise

The claim that the Self-ONN surpasses the CNN baselines rests on the assumption that the different data splitting, normalization, optimizer, and training schedules used in the two studies do not materially affect the comparison; if they do, the 0.010 accuracy gap could be protocol, not architecture.

Editorial extensions

If this is right

  • A Self-ONN with Q=2 offers a drop-in replacement for EfficientNet B1 on this dataset with roughly 22 times fewer parameters and 24 times faster inference, making edge deployment practical.
  • Every Self-ONN variant (Q=2 to Q=5) beats the Q=1 CNN trunk, so on this task adding neuron-level polynomial non-linearity helps more than adding layers.
  • The near-perfect separation of misalignment and broken rotor (zero cross-class errors in the Q=2 confusion matrix) suggests thermal signatures of these two faults are highly distinguishable once the network has enough non-linear capacity.
  • Because the model is shallow and fast, it can be stacked into multi-device monitoring systems where one thermal camera per motor is not affordable.

Reading between the lines

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

  • The accuracy gap over EfficientNet B1 may be partly an artifact of protocol differences: the paper uses stratified folds created without shuffling, per-image min-max normalization, Adam at 0.001, and a different early-stopping schedule than the baseline's AdamW at 0.00003; a matched-protocol rerun is needed to know how much of the 0.010 accuracy difference is due to the Self-ONN itself.
  • The Q parameter looks like a practical tuning dial: rather than scaling network depth or width, an engineer could adjust the Taylor order to trade accuracy against parameter count, and this paper's Q=4 and Q=5 results show the trade-off is not monotonic.
  • The same three-layer Self-ONN could be tested on other rotating-machine thermal datasets (e.g., bearing faults under variable load) to see whether the advantage persists beyond this one motor and camera setup.
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 / 6 minor

Summary. The paper proposes a 2D Self-Organized Operational Neural Network (Self-ONN) with three operational layers for classifying thermal images of squirrel-cage induction motors into healthy, misalignment, and broken-rotor classes. The authors compare their model against CNN baselines from the reference study [5], report that a Self-ONN with Taylor order Q=2 achieves 0.942 ± 0.070 accuracy versus 0.932 ± 0.086 for EfficientNet B1, and highlight reductions in trainable parameters and inference time. The paper also includes a within-paper comparison of Q=1 (a 2D CNN) and Q=2..5 Self-ONNs under a shared protocol, where Q=2 outperforms Q=1 (0.942 versus 0.887).

Significance. If the central comparison were valid, the result would be practically significant: it would show that a shallow, low-parameter operational network can match or exceed much larger CNNs on a thermal-imaging fault-diagnosis task, with clear deployment advantages for edge monitoring. The paper deserves credit for a self-contained within-paper comparison of Q=1 and Q=2..5 models, for reporting per-fold standard deviations, and for quantifying parameter counts and inference durations. However, the headline claim of surpassing EfficientNet B1 depends on cross-protocol comparisons to numbers taken from [5], and the data-splitting procedure raises a leakage concern; these issues make the magnitude and validity of the reported advantage uncertain.

major comments (3)
  1. [Section III.A] The 5-fold split is described as preserving class proportions 'without shuffling the data.' Because the thermal camera records at 4 fps over roughly 30-second sessions, adjacent frames are near-duplicates. Placing contiguous blocks from the same recording session into different folds means that near-identical frames appear on both sides of the train/test boundary, typically inflating cross-validation accuracy. This directly affects the reported 0.942 ± 0.070 figure and the comparison with [5]. The authors should use a session-grouped or temporally separated split and report the resulting accuracy.
  2. [Section IV, Table III] The comparison to the CNN baselines from [5] is uncontrolled. The reference study used random folds, global min-max scaling with dataset-wide extrema, the AdamW optimizer with learning rate 3e-5, and early-stopping patience 10, whereas the present study uses contiguous unshuffled folds, per-image min-max normalization (Equation 6), Adam with learning rate 1e-3, and early-stopping patience 5 with a learning-rate reduction schedule. Any of these differences could contribute to the 0.010 accuracy gap (0.942 vs 0.932), which is smaller than the reported per-fold standard deviations. The claim that the Self-ONN 'surpassed' EfficientNet B1 therefore needs either a matched re-implementation of the baselines under the same protocol or a paired significance test; as written, the claim is unsupported.
  3. [Section IV, Table III] The Taylor order Q is selected after inspecting the test-fold results, with Q=2 reported as the best among Q=1..5. Because Q controls model capacity and multiple Q values are evaluated on the same test folds, the reported 0.942 accuracy is a selected maximum and is likely to be optimistically biased. Q should be selected on validation folds, or the paper should explicitly present all Q results as exploratory and avoid claiming that the best Q value represents expected performance.
minor comments (6)
  1. [Section II] Equation (1) is described as giving 'the input feature map' but x_l_k(m,n) is the pre-activation output of the neuron; the terminology should be corrected to avoid confusion between input and output feature maps.
  2. [Section II, Equation (2)] The domain/codomain notation for the nodal and pool operators is typeset in a way that is hard to parse; please restate the notation with clear dimensions, for example by writing the nodal operator as acting on the K×K patch and the pool operator as reducing that patch to a scalar.
  3. [Conclusion] There is a typo: 'misaligment' should be 'misalignment'.
  4. [Table II] The header 'Standart' should be 'Standard'.
  5. [Introduction] The phrase 'To further the improve performance of CNNs' is ungrammatical; it should read 'To further improve the performance of CNNs'.
  6. [Section IV] The inference-time comparison reports average duration over 100 runs on a single GPU but does not specify batch size, framework, or whether the same software and input pipeline were used for all models; these details should be added for the comparison to be reproducible.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: central claim is an empirical benchmark; Self-ONN equations come from prior published work [15], not from the target results.

full rationale

The paper's central claim is an empirical comparison, not a derivation from first principles. The Self-ONN formulation in Equations (3)-(5) is taken from external prior work [15], and the reported accuracy, loss, etc. are measured results rather than quantities fitted to a target outcome. The comparison to EfficientNet B1 uses externally published baseline numbers from [5]; the headline superiority claim is not constructed from the paper's own inputs. Self-citations [4], [13], [16], [17] appear in the literature review and are not load-bearing for the empirical claim. The 'without shuffling the data' fold construction (Sec III.A), per-image min-max normalization Eq (6), and different optimizer/training schedule relative to [5] are threats to fair comparison and generalization, as is selecting Q=2 after inspecting test-fold results, but these are experimental-design/statistical-reporting issues, not circularity: no reported number is equivalent by construction to fitted inputs. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the authors' prior work, no ansatz is smuggled via citation to force the result. Verdict: no significant circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim relies on the dataset and baseline results from [5], on a particular set of hand-chosen hyperparameters, and on the assumption that the unshuffled fold split does not leak temporal information. No new physical or mathematical entities are introduced. The most consequential free parameter is Q, which is selected on the test set.

free parameters (3)
  • Taylor polynomial order Q = 2 (best on test folds)
    Q is a hyperparameter controlling the degree of the Taylor approximation. The paper reports Q=2 as the best of Q=1..5 based on test-fold accuracy, so the headline result is selected on the test set.
  • Network architecture hyperparameters = 8 filters per layer, kernels 5x5/3x3/2x2, dense 32, tanh
    The small architecture is chosen by hand and is central to the efficiency claim, but no ablation or sensitivity analysis is provided.
  • Training hyperparameters = Adam lr=0.001, batch=16, early-stop patience=5, lr-reduce patience=3, min lr=0.00005
    These differ from the baseline study's choices (AdamW lr=0.00003, batch 8/16, patience 10) and could affect relative performance.
assumptions (4)
  • domain assumption The thermal image dataset and labels from [5] are accurate and the experiments were conducted as described there.
    The paper does not collect new data and provides no independent check of labels, experimental conditions, or image quality; all conclusions rest on the prior study's dataset.
  • domain assumption The baseline CNN results from [5] are directly comparable to the proposed results despite different data splits, normalization, optimizer, and early stopping.
    Table III mixes results from two different training protocols. The paper assumes these differences do not change relative performance, but provides no control experiment or significance test.
  • domain assumption Thermal frames captured in a 30-second sequence can be treated as independent samples for cross-validation.
    Folds were created without shuffling the data (Section III.A). If consecutive frames from the same experiment and load condition appear in both training and test folds, accuracy is optimistically biased.
  • domain assumption The Self-ONN implementation follows [15] and the Taylor expansion in Equation (3) is correctly realized, including the omission of the q=0 constant term.
    No code is provided, so the paper relies on the correctness of the reference implementation. The missing zeroth-order term is a mathematical discrepancy that is not discussed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Thermal Image-based Fault Diagnosis in Induction Machines via Self-Organized Operational Neural Networks." pith.science (2026). https://pith.science/paper/E27M6KAM

@misc{pith2026241205901,
  author       = {Pith},
  title        = {Pith review of: Thermal Image-based Fault Diagnosis in Induction Machines via Self-Organized Operational Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/E27M6KAM}},
  note         = {Machine review of arXiv:2412.05901}
}
read the original abstract

Condition monitoring of induction machines is crucial to prevent costly interruptions and equipment failure. Mechanical faults such as misalignment and rotor issues are among the most common problems encountered in industrial environments. To effectively monitor and detect these faults, a variety of sensors, including accelerometers, current sensors, temperature sensors, and microphones, are employed in the field. As a non-contact alternative, thermal imaging offers a powerful monitoring solution by capturing temperature variations in machines with thermal cameras. In this study, we propose using 2-dimensional Self-Organized Operational Neural Networks (Self-ONNs) to diagnose misalignment and broken rotor faults from thermal images of squirrel-cage induction motors. We evaluate our approach by benchmarking its performance against widely used Convolutional Neural Networks (CNNs), including ResNet, EfficientNet, PP-LCNet, SEMNASNet, and MixNet, using a Workswell InfraRed Camera (WIC). Our results demonstrate that Self-ONNs, with their non-linear neurons and self-organizing capability, achieve diagnostic performance comparable to more complex CNN models while utilizing a shallower architecture with just three operational layers. Its streamlined architecture ensures high performance and is well-suited for deployment on edge devices, enabling its use also in more complex multi-function and/or multi-device monitoring systems.

Figures

Figures reproduced from arXiv: 2412.05901 by the authors.

Figure 1
Figure 1. Proposed 2D Self-ONN architecture with preprocessing steps for fault diagnosis using thermal images from [5]. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Visual representation of the formulations for convolutional, operational, and self-organizing (generative) neurons [18]. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Experimental setup [5]. The experiments were conducted using a 3-phase squirrel cage induction motor, specifically the Celma Indukta Sh 90L-4 model. This motor has a rated power of 1.5 kW, and runs at a speed of 1410 rpm. The experimental setup is depicted in [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Examples of thermal images for a healthy motor, and motors with [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Predictive Failure Detection in Network Hardware Using Thermal Imaging and Deep Learning with Sensor Fusion

    cs.CV 2026-08 reject novelty 4.0 of 10

    A CNN-LSTM model trained on simulated thermal and power data reaches 94% accuracy on a private synthetic dataset, but the practical claim is not supported by real-world validation or a rule-based baseline.

Reference graph

Works this paper leans on

28 extracted references · 22 canonical work pages · cited by 1 Pith paper

  1. [5]

    Unraveling induction motor state through thermal imaging and edge processing: A step towards explainable fault diagnosis,

    M. Piechocki, T. Pajchrowski, M. Kraft, M. Wolkiewicz, and P. Ewert, “Unraveling induction motor state through thermal imaging and edge processing: A step towards explainable fault diagnosis,” Eksploatacja i Niezawodno´s´c – Maintenance and Reliability , vol. 25, no. 3, 2023

    work page 2023

  2. [1]

    A comprehensive review of conventional and intelligence- based approaches for the fault diagnosis and condition monitoring of induction motors,

    R. R. Kumar, M. Andriollo, G. Cirrincione, M. Cirrincione, and A. Tortella, “A comprehensive review of conventional and intelligence- based approaches for the fault diagnosis and condition monitoring of induction motors,” Energies, vol. 15, no. 23, 2022

    work page 2022

  3. [2]

    Implementation of a motor diagnosis system for rotor failure using genetic algorithm and fuzzy classification,

    C.-C. Kuo, C.-H. Liu, H.-C. Chang, and K.-J. Lin, “Implementation of a motor diagnosis system for rotor failure using genetic algorithm and fuzzy classification,” Applied Sciences, vol. 7, p. 31, 12 2016

    work page 2016

  4. [3]

    Rolling element bearing fault diagnosis for rotating machinery using vibration spectrum imaging and convolutional neural networks,

    A. Khodja, G. Noureddine, M. Saadi, and N. Boutasseta, “Rolling element bearing fault diagnosis for rotating machinery using vibration spectrum imaging and convolutional neural networks,” The International Journal of Advanced Manufacturing Technology , vol. 106, pp. 1737– 1751, 01 2020

    work page 2020

  5. [4]

    Smartphone-based bearing fault diagnosis in rotating machinery using audio data and 1d convolu- tional neural networks,

    C. Celebioglu, S. Kilickaya, and L. Eren, “Smartphone-based bearing fault diagnosis in rotating machinery using audio data and 1d convolu- tional neural networks,” in Proceedings of the International Conference on Computer Systems and Technologies 2024 , 2024, pp. 149–154

    2024

  6. [6]

    Diagnosis of the three-phase induction motor using thermal imaging,

    A. Glowacz and Z. Glowacz, “Diagnosis of the three-phase induction motor using thermal imaging,” Infrared Physics & Technology, vol. 81, pp. 7–16, 2017

    work page 2017

  7. [7]

    Three-phase in- duction motor fault detection based on thermal image segmentation,

    A. K. Al-Musawi, F. Anayi, and M. Packianather, “Three-phase in- duction motor fault detection based on thermal image segmentation,” Infrared Physics & Technology , vol. 104, p. 103140, 2020

    work page 2020

  8. [8]

    Thermographic fault diagnosis of electrical faults of com- mutator and induction motors,

    A. Glowacz, “Thermographic fault diagnosis of electrical faults of com- mutator and induction motors,” Engineering Applications of Artificial Intelligence, vol. 121, p. 105962, 2023

    work page 2023

Show all 28 references
  1. [9]

    Rotating machinery fault diagnosis based on convolutional neural network and infrared thermal imaging,

    Y . Li, X. Du, F. Wan, X. Wang, and H. Yu, “Rotating machinery fault diagnosis based on convolutional neural network and infrared thermal imaging,” Chinese Journal of Aeronautics , vol. 33, no. 2, pp. 427–438, 2020

    2020

  2. [10]

    A deep learning approach for electric motor fault diagnosis based on modified inceptionv3,

    L. Xu, S. S. Teoh, and H. Ibrahim, “A deep learning approach for electric motor fault diagnosis based on modified inceptionv3,” Scientific Reports, vol. 14, 05 2024

    2024

  3. [11]

    Fault diagnosis of a rotor-bearing system under variable rotating speeds using two-stage parameter transfer and infrared thermal images,

    H. Shao, W. Li, M. Xia, Y . Zhang, C. Shen, D. Williams, A. Kennedy, and C. W. de Silva, “Fault diagnosis of a rotor-bearing system under variable rotating speeds using two-stage parameter transfer and infrared thermal images,” IEEE Transactions on Instrumentation and Measure-...

    2021

  4. [12]

    Bearing fault diagnosis of induction motor using thermal imaging,

    A. Choudhary, S. Shimi, and A. Akula, “Bearing fault diagnosis of induction motor using thermal imaging,” in 2018 International Confer- ence on Computing, Power and Communication Technologies (GUCON), 2018, pp. 950–955

    2018

  5. [13]

    Real- time motor fault detection by 1-d convolutional neural networks,

    T. Ince, S. Kiranyaz, L. Eren, M. Askar, and M. Gabbouj, “Real- time motor fault detection by 1-d convolutional neural networks,” IEEE Transactions on Industrial Electronics , vol. 63, no. 11, pp. 7067–7075, 2016

    2016

  6. [14]

    An improved fault diagnosis using 1d-convolutional neural network model,

    C.-C. Chen, Z. Liu, G. Yang, C.-C. Wu, and Q. Ye, “An improved fault diagnosis using 1d-convolutional neural network model,” Electronics, vol. 10, no. 1, 2021

    2021

  7. [15]

    Self-organized operational neural networks with gener- ative neurons,

    S. Kiranyaz, J. Malik, H. B. Abdallah, T. Ince, A. Iosifidis, and M. Gabbouj, “Self-organized operational neural networks with gener- ative neurons,” Neural Networks, vol. 140, pp. 294–308, 2021

    2021

  8. [16]

    Improved detection of broken rotor bars by 1-d self-onns,

    L. Eren, O. C. Devecioglu, T. Ince, and M. Askar, “Improved detection of broken rotor bars by 1-d self-onns,” in IECON 2022 – 48th Annual Conference of the IEEE Industrial Electronics Society , 2022, pp. 1–5

    2022

  9. [17]

    Improved domain adaptation approach for bearing fault diagnosis,

    T. Ince, S. Kilickaya, L. Eren, O. C. Devecioglu, S. Kiranyaz, and M. Gabbouj, “Improved domain adaptation approach for bearing fault diagnosis,” in IECON 2022 – 48th Annual Conference of the IEEE Industrial Electronics Society , 2022, pp. 1–6

    2022

  10. [18]

    Self-organized operational neu- ral networks for severe image restoration problems,

    J. Malik, S. Kiranyaz, and M. Gabbouj, “Self-organized operational neu- ral networks for severe image restoration problems,” Neural Networks, vol. 135, pp. 201–211, 2021

    2021

  11. [19]

    Operational neural networks,

    S. Kiranyaz, T. Ince, A. Iosifidis, and M. Gabbouj, “Operational neural networks,” Neural Computing and Applications , vol. 32, no. 11, pp. 6645–6668, 2020

    2020

  12. [20]

    Workswell infrared camera (wic),

    Workswell, “Workswell infrared camera (wic),” 2024. [Online]. Available: https://workswell.eu/thermal-camera-for-production-control/

    2024

  13. [21]

    Pp-lcnet: A lightweight cpu convolutional neural network,

    C. Cui, T. Gao, S. Wei, Y . Du, R. Guo, S. Dong, B. Lu, Y . Zhou, X. Lv, Q. Liu, X. Hu, D. Yu, and Y . Ma, “Pp-lcnet: A lightweight cpu convolutional neural network,” 2021. [Online]. Available: https://arxiv.org/abs/2109.15099

    2021 arXiv

  14. [22]

    Resnet10: A lightweight residual network for remote sensing image classification,

    J. Gong, W. Liu, M. Pei, C. Wu, and L. Guo, “Resnet10: A lightweight residual network for remote sensing image classification,” in 2022 14th International Conference on Measuring Technology and Mechatronics Automation (ICMTMA), 2022, pp. 975–978

    2022

  15. [23]

    Mnasnet: Platform-aware neural architecture search for mobile,

    M. Tan, B. Chen, R. Pang, V . Vasudevan, M. Sandler, A. Howard, and Q. V . Le, “Mnasnet: Platform-aware neural architecture search for mobile,” in 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 2815–2823

    2019

  16. [24]

    Efficientnet: Rethinking model scaling for convolutional neural networks,

    M. Tan and Q. V . Le, “Efficientnet: Rethinking model scaling for convolutional neural networks,” arXiv, 2020

    2020

  17. [25]

    Decoupled weight decay regularization,

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,”

  18. [26]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,”

  19. [2017]

    Available: https://arxiv.org/abs/1412.6980

    [Online]. Available: https://arxiv.org/abs/1412.6980

    arXiv

  20. [2019]

    Available: https://arxiv.org/abs/1711.05101

    [Online]. Available: https://arxiv.org/abs/1711.05101

    arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.