Pith. sign in

REVIEW 5 major objections 5 minor 49 references

Robust Spectral Anomaly Detection in EELS Spectral Images via Three Dimensional Convolutional Variational Autoencoders

T0 review · 5 major / 5 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 A 3D convolutional variational autoencoder trained only on bulk spectra detects subtle spectral anomalies in EELS spectrum images more reliably than PCA, with stable performance across peak-shift magnitudes.

desk verdict The 3D-CVAE formulation is a real novelty for EELS anomaly detection, but the PCA comparison is asymmetric as written and the paper lacks the reproducibility artifacts to back up its central claim. read the letter →

arxiv 2412.16200 v1 pith:NUDPQAMH submitted 2024-12-16 cs.CV cond-mat.mtrl-scics.LG

Seyfal Sultanov , James P Buban , Robert F Klie This is my paper · ORCID
classification cs.CVcond-mat.mtrl-scics.LG
keywords EELSspectrumimagingvariationalautoencoderanomalydetection3Dconvolutionspectralfinestructureprincipalcomponentanalysisunsupervisedlearningtransmissionelectronmicroscopy
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 tries to establish that a three-dimensional convolutional variational autoencoder (3D-CVAE) can detect subtle spectral anomalies in electron energy loss spectroscopy spectrum images (EELS-SI) by learning to reconstruct only the defect-free bulk structure of a material. The authors argue that, unlike PCA, whose linear variance-based decomposition pushes subtle spectral changes into discarded components, the VAE's regularized latent space represents anomalous spectra as close neighbors of their normal counterparts, so the decoder filters out the anomaly and the reconstruction error localizes it. On synthetic Fe L-edge peak shifts injected into BiFeO3 data, the VAE produces a bimodal error distribution that cleanly separates anomalous from normal pixels, with only 6 of 38,000 spectra misclassified in the shown example, and its F1 score stays high across shift magnitudes while PCA's fluctuates. The motivation is that core-loss near-edge fine structure encodes valence changes, defects, and interface effects that matter for understanding complex materials, so an unsupervised, element-agnostic detector that works on full datacubes would be broadly useful.

What carries the argument

The central object is the three-dimensional convolutional variational autoencoder (3D-CVAE), a VAE whose encoder and decoder use 3D convolutions over the datacube's two spatial axes and one energy-loss axis, trained with a cross-entropy (negative log-likelihood) reconstruction loss plus a KL-divergence term with weighting $\beta = 1.2$. The cross-entropy loss treats each energy channel of a normalized spectrum as a discrete class, matching the counting nature of electron energy-loss events. The trained model is applied to shards of the datacube, and for each pixel the Pearson correlation coefficient between the input and reconstructed spectrum in the Fe L-edge window (690-730 eV) serves as the anomaly score; Otsu's method with a unimodality check converts the PCC histogram into anomaly labels. The load-bearing mechanism is the KL-regularized latent space: because anomalous spectra are encoded near their normal counterparts, the decoder outputs bulk-like features and the reconstruction residual localizes the defect.

What would settle it

Take a real EELS-SI dataset containing independently verified atomic-scale defects (for example, a known grain boundary or oxygen vacancy column with spectra confirmed by atomic-resolution imaging or simulation) and run the same 3D-CVAE trained only on bulk regions. If the reconstruction-error histogram is not bimodal or the F1 against the verified defect map is near chance while PCA does no worse, the central claim fails. A cheaper falsifier: show that the bimodal separation disappears when the injected anomaly shape is changed to a realistic defect signature (broadening, edge-onset shift, or intensity change) rather than a rigid 2.5 eV shift.

Watch

Extended reading notes

Core claim

The central claim is that a VAE trained only on bulk spectra and applied to the full three-dimensional EELS datacube yields a Pearson-correlation error map whose histogram is bimodal: a high-correlation bulk population and a low-correlation anomalous population, separable by Otsu's threshold. The encoder places anomalous and normal variants of the same spectral region close together in the 40-dimensional latent space, so the decoder's reconstruction reverts toward bulk and the remaining discrepancy marks the anomaly. Across injected peak shifts, the VAE keeps high F1 with high precision and recall, while PCA with 3, 4, or 5 components shows periodic fluctuations that the paper attributes to alignment between the shift direction and the PCA basis. The performance gap narrows as anomaly concentration decreases, but the VAE still maintains reconstruction quality in low signal-to-noise regions such as the O K edge.

Load-bearing premise

The evaluation assumes that synthetic 2.5 eV Fe L-edge peak shifts injected into spatially clustered pixels faithfully mimic real defect-induced spectral changes in materials; all quantitative claims about detecting defects rest on this proxy, and no real defect spectra are tested.

Editorial extensions

If this is right

  • EELS-SI anomaly detection becomes fully unsupervised: training needs only bulk spectra, and no labels, feature engineering, or prior knowledge of the material's elements are required.
  • The method detects subtle core-loss peak shifts with a consistent F1 across shift magnitudes, so one threshold works for weak and strong anomalies alike, whereas PCA's threshold would need retuning per shift size.
  • The approach works on the full three-dimensional datacube rather than individual spectra, preserving spatial context that helps distinguish clustered defects from noise.
  • The performance advantage over PCA shrinks as the fraction of anomalous spectra decreases, so for very sparse defects PCA remains competitive.
  • The latent-space analysis indicates the encoder treats anomalous spectra as variants of bulk spectra, which is why reconstruction filters anomalies rather than memorizing them.

Reading between the lines

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

  • If the bimodal error separation holds on real specimens, the same trained-on-bulk pipeline could be used for on-the-fly defect mapping during STEM-EELS acquisition, flagging regions for longer spectral acquisition; the paper does not discuss this use.
  • The periodic PCA failures suggest a geometric cause beyond the paper's framing: a linear subspace can only represent shifts along its retained basis, so shifts nearly orthogonal to the principal components become invisible; a nonlinear generative model does not share that blind direction. This explanation is an inference, not stated as a result.
  • A natural, untested extension is to replace cross-entropy with a Poisson negative log-likelihood, which is the actual noise model of electron counting; if it performs comparably or better, it would strengthen the claim that the choice of reconstruction loss is what drives the bimodality.
  • The synthetic 2.5 eV rigid shift is a single proxy; testing the method on real defect signatures (edge broadening, white-line ratio changes, or onset shifts) would tell whether the advantage transfers to the defects that actually occur in materials.
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

5 major / 5 minor

Summary. The paper proposes a 3D convolutional variational autoencoder (3D-CVAE) for anomaly detection in EELS spectrum-image datacubes. The model is trained on bulk spectra using a cross-entropy reconstruction term plus a KL regularization term, and anomalies are scored by the Pearson correlation coefficient between input and reconstructed spectra, with Otsu thresholding for classification. The method is evaluated on synthetic Fe L-edge peak-shift anomalies injected into EELS-SI data from BiFeO3/SrTiO3, and compared against PCA with 3, 4, and 5 components. The authors report that the VAE achieves high F1 scores across shift magnitudes, shows bimodal separation of normal and anomalous pixels, and misclassified only 6 of 38,000 spectra in a specific example, while PCA exhibits periodic fluctuations.

Significance. If the evaluation is sound, the paper is a useful methodological contribution to unsupervised EELS-SI analysis: 3D convolutions are a natural fit for the spatio-spectral structure of datacubes, the use of a probabilistic reconstruction error is well motivated, and the reported '6 of 38,000' misclassified spectra is a concrete, checkable quantitative claim. The clear synthetic benchmark, the explicit comparison against PCA, and the latent-space analysis are strengths. However, the significance is contingent on two load-bearing points: the fairness of the PCA baseline and the reproducibility of the experiments. The paper does not provide code, data, error bars, or a complete architecture description, which limits the confidence one can place in the headline comparison.

major comments (5)
  1. [Results, Figure 4 and Methods] The central F1 comparison is undermined by a potential asymmetry in how the two methods are trained. The Methods section never states how the PCA components are obtained; the Results say 'the analysis pipeline processes the anomaly-injected datacube through both methods' and that PCA 'is optimal when anomalies are small in number and sparsely distributed.' Both statements are consistent with PCA being fit to the full anomaly-injected test datacube, whereas the VAE is trained exclusively on bulk spectra. If so, PCA is given contaminated training data, and the reported F1 gap, periodic fluctuations, and diminishing advantage with anomaly concentration could be artifacts of the protocol rather than intrinsic properties of the methods. Please state explicitly whether PCA components are computed on a clean training split or on the contaminated test cube; if the latter, rerun the comparison with PCA fit to the same clean bulk shards used for VAE training.
  2. [Results, Figure 4] All F1 comparisons are reported without error bars, confidence intervals, or repeated-seed statistics. Since VAE training is stochastic and the anomaly injections are synthetic, a single run cannot support the claim of 'consistent performance across various shift magnitudes.' Please report mean and standard deviation over at least 5-10 independent training runs and multiple anomaly-injection realizations.
  3. [Methods, Figure 7 and Eqs. (1)-(3)] The manuscript provides no complete architecture or training configuration. Figure 7 is only a schematic; the text gives no layer counts, kernel sizes, strides, channels, latent dimension J (40 appears only later in the latent-space analysis), shard size L, optimizer, learning rate, batch size, or epoch count. Eqs. (1)-(3) define the loss, but the hyperparameter choices (β=1.2, J=40, 24×24×L shards, PCA component counts) are not backed by a described search procedure or a validation split. Without these details, the experiments cannot be reproduced or the robustness of the tuned values assessed.
  4. [Results, Figure 2 and Conclusion] The entire empirical case rests on synthetic Fe L-edge peak shifts injected into spatially clustered pixels. The claim that 2.5 eV shifts 'represent realistic defect-induced changes' is asserted without supporting evidence or reference, and no real defect spectra are tested. The practical conclusion that the method is 'particularly valuable for analyzing complex material systems' therefore goes beyond the demonstrated scope. Please either provide validation on real defect-containing EELS-SI data or substantially temper the generalization claim and explicitly frame the contribution as a synthetic-benchmark evaluation.
  5. [Results, Figure 5] The latent-space analysis reports cosine similarities between encodings of normal and anomaly-injected shard pairs, but it does not quantify the detection benefit of this proximity or compare it against the PCA latent space. The claim that 'lower dimensional representations are robust to anomalies' is not directly supported by the presented metric. Please provide a quantitative comparison, such as the distribution of off-diagonal cosine similarities or a reconstruction-error comparison at fixed latent dimension.
minor comments (5)
  1. [Abstract vs. Methods, Eq. (1)] The abstract says the model employs 'negative log-likelihood loss,' but Eq. (1) defines a cross-entropy loss; please clarify that for normalized spectra the cross-entropy is equivalent to the negative log-likelihood of a categorical distribution, or correct the terminology.
  2. [Methods, Eq. (3)] There is a typographical error in the sentence preceding Eq. (3): 'ensuring that it is smooth and aligned with a prior Gaussian distribution,.' contains an extra comma and period.
  3. [Results, '6 of 38,000'] The statement that 'our VAE approach achieved high classification accuracy, with only 6 anomalous individual spectra misclassified as part of the bulk material structure out of 38,000 total spectra' needs context: how many anomalous spectra were injected in total, and what were the corresponding precision, recall, and F1 for this example?
  4. [References] References [38]-[46] do not appear to be cited in the text, and references [42] and [46] are duplicate entries for the same paper. Please remove uncited references and merge duplicates.
  5. [Figure 5] Please state whether the 64 shard pairs used for the cosine-similarity analysis come from the training split or the test split, and how the 'close proximity' is quantified beyond visual inspection of the diagonal.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation chain: the VAE anomaly-detection result is an empirical reconstruction-error comparison, and the PCA protocol concern is a fairness issue rather than a circularity.

full rationale

The paper's claimed result is not a formal derivation but an empirical protocol: train a 3D-CVAE on bulk EELS-SI spectra (Eqs. 1-3), reconstruct full datacubes, and score anomalies by the Pearson correlation between original and reconstructed spectra (Eq. 4). Nothing in these equations defines the anomaly score in terms of the injected labels, the PCA baseline, or the reported F1 values; the result could fail (a flexible VAE might reconstruct shifted spectra well), so it is not forced by construction. The self-citations [30,31] describe the authors' earlier CVAE defect-detection work on STEM images and are contextual, not load-bearing; the present EELS-SI claim is tested on new data and does not depend on those citations as proof. The hyperparameter choices (beta = 1.2, 40 latent dimensions) are stated as tuned, but the manuscript does not say they were tuned against the test anomalies, so this is at most a reproducibility or overfitting concern, not a demonstrated circular step. The PCA comparison is under-specified: the text says 'The analysis pipeline processes the anomaly-injected datacube through both methods' (Results) and never states whether PCA components were fitted on clean bulk data or on the contaminated cube. If PCA was fitted on the contaminated cube, the comparison is asymmetric and the claimed superiority would be weakened; however, that is an experimental-design confound, not an equivalence of the VAE's prediction to its inputs by construction. The paper also acknowledges its own limitation that quantitative metrics for low-SNR regions are not established (Conclusion). Overall, this is an honest non-finding: no circular step is exhibited by the paper's own equations or explicit protocol.

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

The method relies on standard VAE machinery plus two domain assumptions: that latent-space reconstruction filters anomalies, and that categorical cross-entropy is appropriate for EELS counts. The only ad hoc element is the synthetic peak-shift proxy for real defects. No new physical entities are introduced. All central evidence depends on four free parameters that are tuned or chosen by the authors without sensitivity analysis.

free parameters (4)
  • beta (KL weight) = 1.2
    Chosen by hyperparameter tuning to balance reconstruction and latent-space regularization; directly affects detection quality.
  • Latent dimension J = 40
    Fixed dimensionality of the VAE latent space; no sensitivity analysis is reported.
  • Shard size = 24x24xL
    The datacube is segmented into 24 by 24 spatial tiles before processing; the choice affects spatial context available to the model.
  • PCA component count = 3, 4, 5
    The number of PCA components is a free axis; the best variant among those tested is used for comparison.
assumptions (3)
  • domain assumption A VAE trained on normal data will map anomalous inputs near their normal counterparts in latent space, causing reconstruction to filter out anomalous features.
    Methods section states that 'the encoder recognizes anomalous and normal spectra as fundamentally the same data point, leading to reconstructions that effectively filter out the anomalous features.' This is assumed, not derived.
  • domain assumption Normalized EELS intensities can be treated as event probabilities, making categorical cross-entropy a valid reconstruction loss.
    Equation (2) treats each energy channel as a class and normalized intensities as probabilities of electron energy loss events. This is plausible but not justified with detector statistics.
  • ad hoc to paper Injected Fe L-edge peak shifts represent realistic defect-induced spectral changes.
    Results section says the 2.5 eV shift was 'chosen to represent realistic defect-induced changes in electronic structure', but no real defect spectra are tested.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Robust Spectral Anomaly Detection in EELS Spectral Images via Three Dimensional Convolutional Variational Autoencoders." pith.science (2026). https://pith.science/paper/NUDPQAMH

@misc{pith2026241216200,
  author       = {Pith},
  title        = {Pith review of: Robust Spectral Anomaly Detection in EELS Spectral Images via Three Dimensional Convolutional Variational Autoencoders},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NUDPQAMH}},
  note         = {Machine review of arXiv:2412.16200}
}
read the original abstract

We introduce a Three-Dimensional Convolutional Variational Autoencoder (3D-CVAE) for automated anomaly detection in Electron Energy Loss Spectroscopy Spectrum Imaging (EELS-SI) data. Our approach leverages the full three-dimensional structure of EELS-SI data to detect subtle spectral anomalies while preserving both spatial and spectral correlations across the datacube. By employing negative log-likelihood loss and training on bulk spectra, the model learns to reconstruct bulk features characteristic of the defect-free material. In exploring methods for anomaly detection, we evaluated both our 3D-CVAE approach and Principal Component Analysis (PCA), testing their performance using Fe L-edge peak shifts designed to simulate material defects. Our results show that 3D-CVAE achieves superior anomaly detection and maintains consistent performance across various shift magnitudes. The method demonstrates clear bimodal separation between normal and anomalous spectra, enabling reliable classification. Further analysis verifies that lower dimensional representations are robust to anomalies in the data. While performance advantages over PCA diminish with decreasing anomaly concentration, our method maintains high reconstruction quality even in challenging, noise-dominated spectral regions. This approach provides a robust framework for unsupervised automated detection of spectral anomalies in EELS-SI data, particularly valuable for analyzing complex material systems.

Figures

Figures reproduced from arXiv: 2412.16200 by the authors.

Figure 1
Figure 1. FIG. 1. Comparison of VAE and PCA reconstructions and their anomaly detection performance. (a) Split visualization of [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Example of an injected peak shift anomaly in EELS [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. FIG. 4. Performance comparison across different magnitudes [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figures from the paper (4 more)
Figure 3
Figure 3. Figure 3: FIG. 3. Distribution of pixels across Pearson Correlation Co [PITH_FULL_IMAGE:figures/full_fig_p005_3.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Visualization of latent space relationships through co [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6. Precision-recall curves comparing anomaly detection performance between our 3D-CVAE model and PCA with 3,4,5 [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7. Schematic representation of the 3D-CVAE architecture. The encoder processes the input EELS SI datacube through [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 43 canonical work pages

  1. [1]

    L. Song, L. J. Ci, H. Lu, P. B. Sorokin, C. H. Jin, J. Ni, A. G. Kvashnin, D. G. Kvashnin, J. Lou, B. I. Yakobson, and P. M. Ajayan, Nano Letters 10, 3209 (2010)

    work page 2010

  2. [2]

    R. F. Klie, J. P. Buban, M. Varela, A. Franceschetti, C. Jooss, Y. Zhu, N. D. Browning, S. T. Pantelides, and S. J. Pennycook, Nature 435, 475 (2005)

    work page 2005

  3. [3]

    and catalysts. [4] A particularly powerful approach to materials characterization is the combination of scan- ning transmission electron microscopy (STEM) [5] with electron energy-loss spectroscopy (EELS), [6] which can measure the local density of states up to single atomic- column resolutions. [7] This approach is often referred to as EELS spectrum imag...

    work page Pith review arXiv 2024

  4. [4]

    Through hy- perparameter tuning, we determined β = 1.2 to provide the optimal balance between latent space organization and reconstruction quality for this specific dataset

    enforce stricter regularization at the cost of recon- struction accuracy, while lower values prioritize precise reconstructions of the input spectra [21]. Through hy- perparameter tuning, we determined β = 1.2 to provide the optimal balance between latent space organization and reconstruction quality for this specific dataset. Results—To validate our appr...

    work page

  5. [5]

    S. J. Pennycook and L. A. Boatner, Nature 336, 565 (1988)

    work page 1988

  6. [6]

    P. M. Voyles, D. A. Muller, J. L. Grazul, P. H. Citrin, and H. J. L. Gossmann, Nature 416, 826 (2002)

    work page 2002

  7. [7]

    K. Sun, J. Liu, N. Nag, and N. D. Browning, Catalysis Letters 84, 193 (2002)

    work page 2002

  8. [8]

    Jeanguillaume and C

    C. Jeanguillaume and C. Colliex, Ultramicroscopy 28, 252 (1989)

    work page 1989

Show all 49 references
  1. [9]

    Egerton, Electron Energy Loss Spectroscopy in the Electron Microscope, 2nd ed

    R. Egerton, Electron Energy Loss Spectroscopy in the Electron Microscope, 2nd ed. (Springer Science & Busi- ness Media, New York, 2011)

    2011

  2. [10]

    Varela, S

    M. Varela, S. D. Findlay, A. R. Lupini, H. M. Christen, A. Y. Borisevich, N. Dellby, O. L. Krivanek, P. D. Nellist, M. P. Oxley, L. J. Allen, and S. J. Pennycook, Physical Review Letters 92, 095502 (2004)

    2004

  3. [11]

    M. D. Li, C. Z. Chang, L. J. Wu, J. Tao, W. W. Zhao, M. H. W. Chan, J. S. Moodera, J. Li, and Y. M. Zhu, Physical Review Letters 114 (2015), 10.1103/Phys- RevLett.114.146802

    2015 doi

  4. [12]

    N. D. Browning, M. F. Chisholm, S. J. Pennycook, D. P. Norton, and D. H. Lowndes, Physica C 212, 185 (1993)

    1993

  5. [13]

    R. F. Klie, J. C. Zheng, Y. Zhu, M. Varela, J. Wu, and C. Leighton, Physical Review Letters 99, 047203 (2007)

    2007

  6. [14]

    M. H. Mobarak, M. A. Mimona, M. A. Islam, N. Hossain, F. T. Zohura, I. Imtiaz, and M. I. H. Rimon, Applied Surface Science Advances 18, 100523 (2023)

    2023

  7. [15]

    R. F. Klie and N. D. Browning, Applied Physics Letters 77, 3737 (2000)

    2000

  8. [16]

    Lagunas, D

    F. Lagunas, D. Bugallo, F. Karimi, Y. J. Yang, H. O. Badr, J. H. Cope, E. Ferral, M. W. Barsoum, Y. J. Hu, and R. F. Klie, Chemistry of Materials 36, 2743 (2024)

    2024

  9. [17]

    R. Lin, R. Zhang, C. Wang, X. Q. Yang, and H. L. Xin, Scientific Reports 11 (2021), 10.1038/s41598-021-84499- w

    2021 doi

  10. [18]

    LeCun, Y

    Y. LeCun, Y. Bengio, and G. Hinton, Nature 521, 436 (2015)

    2015

  11. [19]

    M. I. Jordan and T. M. Mitchell, Science 349, 255 (2015)

    2015

  12. [20]

    G. E. Hinton and R. R. Salakhutdinov, Science 313, 504 (2006)

    2006

  13. [21]

    Computer vision-aided atom track- ing in stem imaging,

    Y. Hui and Y. Liu, “Computer vision-aided atom track- ing in stem imaging,” (2018), arXiv:1809.05076 [cs.CV]

    2018 arXiv

  14. [22]

    Somnath, C

    S. Somnath, C. R. Smith, S. V. Kalinin, M. Chi, A. Borisevich, N. Cross, G. Duscher, and S. Jesse, Advanced Structural and Chemical Imaging 4 (2018), 10.1186/s40679-018-0052-y

    2018 doi

  15. [23]

    S. Mei, J. Ji, Y. Geng, Z. Zhang, X. Li, and Q. Du, IEEE Transactions on Geoscience and Remote Sensing 57, 6808 (2019)

    2019

  16. [24]

    Auto-encoding varia- tional bayes,

    D. P. Kingma and M. Welling, “Auto-encoding varia- tional bayes,” (2022), arXiv:1312.6114 [stat.ML]. 7

    2022 arXiv

  17. [25]

    Carleo, I

    G. Carleo, I. Cirac, K. Cranmer, L. Daudet, M. Schuld, N. Tishby, L. Vogt-Maranto, and L. Zdeborov´ a, Reviews of Modern Physics 91, 045002 (2019)

    2019

  18. [26]

    C. M. Pate, J. L. Hart, and M. L. Taheri, Scientific Reports 11, 19515 (2021)

    2021

  19. [27]

    Ziatdinov, C

    M. Ziatdinov, C. Y. T. Wong, and S. V. Kalinin, Machine Learning: Science and Technology 4 (2023), 10.1088/2632-2153/ad073b

    2023 doi

  20. [28]

    Ziatdinov, A

    M. Ziatdinov, A. Ghosh, T. Wong, and S. V. Kalinin, Nature Machine Intelligence 4, 1101 (2022)

    2022

  21. [29]

    Yamaguchi, Y

    H. Yamaguchi, Y. Hashimoto, G. Sugihara, J. Miy- ata, T. Murai, H. Takahashi, M. Honda, A. Hishimoto, and Y. Yamashita, Frontiers in Neuroscience 15 (2021), 10.3389/fnins.2021.652987

    2021

  22. [30]

    X. Shu, T. Bao, Y. Zhou, R. Xu, Y. Li, and K. Zhang, Structural Health Monitoring 22, 39 (2023)

    2023

  23. [31]

    Y. Fan, G. Wen, D. Li, S. Qiu, M. D. Levine, and F. Xiao, Computer Vision and Image Understanding 195, 102920 (2020)

    2020

  24. [32]

    Lecun, L

    Y. Lecun, L. Bottou, Y. Bengio, and P. Haffner, Pro- ceedings of the IEEE 86, 2278 (1998)

    1998

  25. [33]

    The CE Loss quantifies the discrepancy between the original spectra and their recon- structions

    is a weighting factor that controls the trade-off be- tween the reconstruction accuracy (governed by the CE Loss) and the regularization of the latent space (enforced by the KL divergence term). The CE Loss quantifies the discrepancy between the original spectra and their reco...

  26. [34]

    Prifti, J

    E. Prifti, J. P. Buban, A. S. Thind, and R. F. Klie, Small 19 (2023), 10.1002/smll.202205977

    2023 doi

  27. [35]

    R. A. W. Ayyubi, J. P. Buban, and R. F. Klie, Microscopy and Microanalysis 30 (2024), 10.1093/mam/ozae044.180

    2024 doi

  28. [36]

    Higgins, L

    I. Higgins, L. Matthey, A. Pal, C. Burgess, X. Glorot, M. Botvinick, S. Mohamed, and A. Lerchner, in Inter- national Conference on Learning Representations(2017)

    2017

  29. [37]

    Pearson and F

    K. Pearson and F. Galton, Proceedings of the Royal So- ciety of London 58, 240 (1895)

  30. [38]

    Otsu, IEEE Transactions on Systems, Man, and Cy- bernetics 9, 62 (1979)

    N. Otsu, IEEE Transactions on Systems, Man, and Cy- bernetics 9, 62 (1979)

    1979

  31. [39]

    Van Rijsbergen, in Proceedings of the joint IBM/University of Newcastle upon tyne seminar on data base systems, Vol

    C. Van Rijsbergen, in Proceedings of the joint IBM/University of Newcastle upon tyne seminar on data base systems, Vol. 79 (1979) pp. 1–14

    1979

  32. [40]

    High-resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High-resolution image synthesis with latent diffusion models,” (2022), arXiv:2112.10752 [cs.CV]

    2022 arXiv

  33. [41]

    Biswas, M

    A. Biswas, M. Ziatdinov, and S. V. Kalinin, Machine Learning: Science and Technology 4, 045004 (2023)

    2023

  34. [42]

    J. Sun, X. Wang, N. Xiong, and J. Shao, IEEE Access 6, 33353 (2018)

    2018

  35. [43]

    Pytorch: An imperative style, high-performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. K¨ opf, E. Yang, Z. DeVito, M. Rai- son, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala, “Pytorch: An imperative style, hig...

    2019 arXiv

  36. [44]

    S. J. Pennycook and P. D. Nellist, eds., Scanning Trans- mission Electron Microscopy(Springer New York, 2011)

    2011

  37. [45]

    Variational autoencoder based anomaly detection using reconstruction probability,

    J. An and S. Cho, “Variational autoencoder based anomaly detection using reconstruction probability,” (2015)

    2015

  38. [46]

    Trans- form invariant auto-encoder,

    T. Matsuo, H. Fukuhara, and N. Shimada, “Trans- form invariant auto-encoder,” (2017), arXiv:1709.03754 [cs.CV]

    2017 arXiv

  39. [47]

    Ng and M.-F

    K.-K. Ng and M.-F. Yang, Physical Review B 108, 214428 (2023)

    2023

  40. [48]

    Cheng, E

    Z. Cheng, E. Zhu, S. Wang, P. Zhang, and W. Li, IEEE Access 9, 43991 (2021)

    2021

  41. [49]

    Variational autoencoder based anomaly detection using reconstruction probability,

    J. An and S. Cho, “Variational autoencoder based anomaly detection using reconstruction probability,” (2015). 8 END MA TTER FIG. 6. Precision-recall curves comparing anomaly detection performance between our 3D-CV AE model and PCA with 3,4,5 components. The V AE demonstrates c...

    2015

Pith tools

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