Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

phaser: A unified and extensible framework for fast electron ptychography

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper presents phaser, an open-source Python package that lets users specify electron ptychography reconstructions as declarative plans and run them through conventional or gradient-descent engines on multiple backends, claiming a…

desk verdict A solid, genuinely useful software paper whose headline speedup is plausible but underdocumented; the benchmark needs to close precision, tuning, and data-availability gaps. read the letter →

arxiv 2505.14372 v2 pith:5LAIBZUM submitted 2025-05-20 physics.comp-ph cond-mat.mtrl-sci

Colin Gilgenbach , Menglin Zhu , James M. LeBeau This is my paper · ORCID
classification physics.comp-phcond-mat.mtrl-sci
keywords electronptychographyscanningtransmissionmicroscopymultislicegradientdescentreconstructionJAXePIELSQMLregularization
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

Electron ptychography reconstructs a specimen's structure from many overlapping diffraction patterns, and multislice ptychography extends this to thick samples, but reconstruction software is often specialized and slow to iterate on. The paper introduces phaser, an open-source package that unifies the two main algorithmic families—traditional iterative engines like ePIE and LSQML, and gradient-descent optimization with automatic differentiation—behind a single declarative reconstruction-plan format. Its central performance claim is that, with the JAX backend, the inner reconstruction loop is compiled into a fast GPU kernel, reaching under three seconds per iteration on a 20-slice multislice dataset and running about six times faster than fold_slice/PtychoShelves. The paper also demonstrates reconstructions of experimental datasets and a systematic study of how regularization choices affect convergence and image quality. If the claims hold, phaser gives materials scientists a practical platform for faster, more flexible ptychographic reconstruction.

What carries the argument

The central object is the reconstruction plan, a declarative YAML or JSON file that specifies data loading, preprocessing, a sequence of reconstruction engines, noise models, regularizers, and user-defined hooks. The mechanism carrying the speed argument is JAX's just-in-time compilation of the inner loop that runs the forward model—multislice propagation of probe modes through object slices followed by a Fourier transform to detector intensities—and the corresponding update step. Grouping and shuffling probe positions makes the update stochastic, and both the analytical wavefront updates of ePIE and LSQML and the autodifferentiated gradients of the gradient-descent engine are expressed through the same modular hook interface.

What would settle it

Run fold_slice/PtychoShelves' LSQML engine on the same RTX 3080 and dataset with its own hyperparameters optimized (for example, with Optuna) rather than fixed to the reference values, and measure seconds per iteration and final RMS error; if the per-iteration time ratio falls below sixfold, or the reconstruction-quality gap reverses or disappears, the paper's central performance and quality claims would be weakened.

Watch

Extended reading notes

Core claim

The authors claim that one software framework can combine conventional ptychographic algorithms and gradient-descent-based algorithms without sacrificing speed, and they attribute the speed gain to architecture: the per-group loop over probe positions is the bottleneck, and just-in-time compiling that loop with JAX removes Python interpreter overhead that dominates in interpreter-based packages. The reported result is a six-fold improvement in seconds per iteration compared with fold_slice/PtychoShelves, down to less than 3 s/iter for 6400 probe positions across 20 slices. On a benchmark PrScO3 dataset, both LSQML and gradient descent outperform ePIE, with gradient descent giving slightly better separation of atomic dumbbells. The paper further reports that regularization parameters act in two distinct ways—some mainly change convergence rate, while others mainly affect final resolution and contrast—and that, in a simulated silicon crystal, individual Sn interstitials are located along the depth axis with an RMS error of 1.1 Å and a depth resolution of about 1.9 nm.

Load-bearing premise

The central performance claim assumes that seconds per iteration measured on one RTX 3080 in a virtual machine with equivalent reconstruction parameters is a fair and representative comparison, and that it is fair to compare Optuna-tuned phaser reconstructions against fold_slice/PtychoShelves run with its reference fixed parameters; if either condition is not representative, the reported speed and quality advantages could shrink.

Editorial extensions

If this is right

  • If the benchmark transfers to other machines, JAX-backed phaser would let users iterate on multislice ptychography reconstructions in minutes rather than hours, making parameter exploration practical.
  • Because engines share one plan format and hook interface, users can switch between ePIE, LSQML, and gradient descent on the same dataset without rewriting data pipelines, simplifying algorithmic comparison.
  • The reported 1.1 Å depth precision for single Sn interstitials in silicon suggests the framework can locate individual dopant atoms along the beam direction, supporting three-dimensional materials characterization.
  • The regularization study implies that phaser reconstructions require parameter tuning, and the built-in coupling to the Optuna hyperparameter optimizer provides an automated path to finding good parameters.
  • The client-server architecture allows reconstructions to run on local GPUs, clusters, or cloud workers while being viewed live, which could enable in-microscope checks of data quality during acquisition.

Reading between the lines

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

  • The six-fold speed advantage is measured against fold_slice/PtychoShelves using that package's published reconstruction parameters; if that package were also hyperparameter-tuned, the quality gap on the PrScO3 dataset could narrow, though the per-iteration speed advantage would likely persist because it comes from just-in-time compilation rather than optimizer choice.
  • A natural extension is to test whether the speedup holds on newer GPUs, multi-GPU setups, or reduced-precision arithmetic, which the authors flag as future work and which machine-learning hardware trends make plausible.
  • Because the noise model is a modular hook feeding the loss, the same infrastructure could be used to benchmark alternative detector noise models or forward models under identical conditions, without changing the reconstruction engine.
  • The versioned, declarative plan format could serve as machine-readable provenance for publications, making reconstruction parameters reproducible and portable across laboratories.
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 / 4 minor

Summary. The paper introduces phaser, an open-source Python package for electron ptychography. It combines a declarative reconstruction-plan interface, conventional engines (ePIE, LSQML) and a gradient-descent engine built on autodifferentiation, multiple computational backends (NumPy, CuPy, JAX), mixed-state probes, probe-position correction, multislice forward models, and several built-in regularizers. The headline claim is that the JAX backend achieves roughly a six-fold reduction in seconds-per-iteration relative to fold_slice/PtychoShelves on a multislice benchmark. The paper also reports successful reconstructions of experimental datasets (PrScO3, BaTiO3, Si) and a depth-sensitivity analysis of simulated Sn interstitials in Si.

Significance. If the performance claim is robust under controlled conditions, phaser is a valuable community resource: it is one of the first packages to combine a unified declarative interface with JAX-based JIT-compiled multislice reconstruction, and the open-source license, versioned plan schema, and hyperparameter-optimization integration are concrete reproducibility features. The experimental demonstrations on multiple datasets and the quantified dopant-depth analysis are strengths. However, the central speedup claim is currently supported by a benchmark that does not fully control precision, tuning, and reproducibility variables, so the paper's significance as a performance contribution depends on tightening the comparison.

major comments (4)
  1. [Sec. VI (Methods) and Sec. IV.A] The performance benchmark does not state the numerical precision used for either phaser or fold_slice/PtychoShelves. Since phaser defaults to single precision (Sec. II.B) and MATLAB typically operates in double precision, a precision mismatch alone could account for a substantial part of the reported speed ratio on a consumer GPU. Please report the precision settings for both codes and, ideally, include a matched-precision comparison so that the speedup can be attributed to the JAX backend rather than to dtype.
  2. [Abstract and Sec. V (Conclusions)] The abstract and conclusions state a 'six-fold' improvement in iteration speed without the qualification given in Sec. IV.A, where the 5-6x advantage is described as 'most stark at small groupings.' Figure 4 shows a strong dependence on grouping; the headline claim should be accompanied by the grouping size and other conditions, and the abstract and conclusions should be worded to match the evidence.
  3. [Sec. IV.A] The paper states that performance must be benchmarked 'including the speed of each iteration as well as the total time to convergence,' but only per-iteration times are reported. In addition, the quality comparison of Fig. 5 is asymmetric: phaser engines were tuned with Optuna, while fold_slice used fixed parameters from Chen et al. [3] (Sec. VI). This confounds both time-to-convergence and reconstruction-quality conclusions. Please either include a matched time-to-convergence benchmark with comparable tuning effort, or explicitly restrict the claim to per-iteration speed.
  4. [Sec. VI and Code & Data availability] The Dryad repository is referenced as '[link]' and the benchmark descriptions provide no run-to-run variability (a single run per configuration on one RTX 3080 in a VM). For a headline speed claim, the repository link must be populated with data and reconstruction plans, and the benchmark should include at least several runs per configuration with reported mean and spread, plus software versions and environment details.
minor comments (4)
  1. [Throughout] The package name is written inconsistently as 'fold slice/PtychoShelves' and 'fold_slice/PtychoShelves'; please standardize to the project's actual name and clarify that the comparison used the fold_slice fork of PtychoShelves.
  2. [Sec. VI] 'a Nvidia RTX 3080' should read 'an Nvidia RTX 3080,' and the specific MATLAB, CUDA, and driver versions used should be listed.
  3. [Sec. IV.B] The sentence referencing 'Figure 7e' for the 'final reconstructed object with varying object L2 and object Tikhonov' would benefit from a more explicit pointer to the panel layout, since Figure 7 is a multi-panel composite.
  4. [Code & Data availability] The placeholder '[link]' for the Dryad record should be replaced with the actual DOI before publication.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: phaser's claims are benchmarked against external software and external experimental datasets, with no fitted parameter renamed as a prediction and no load-bearing self-citation chain.

full rationale

The paper's central claims are software-engineering and empirical-benchmark claims: reconstruction speed, reconstruction quality on experimental datasets, and regularization behavior. The speed claim is supported by a direct comparison against an external package, fold_slice/PtychoShelves (ref. 39), using equivalent reconstruction parameters and a stated hardware configuration, and the quality comparison uses the external PrScO3 dataset from Chen et al. (ref. 3) and a simulated Sn-in-Si test with known ground truth. No physical prediction is derived from a constant fitted to the same data. The self-citations (refs. 7 and 43) supply prior methodological context and the Si experimental dataset, but they do not carry the load of the speed or quality claims: the 6x figure is measured against external software, not derived from a self-cited theorem. The inverse-problem uniqueness statement is cited to the external work of Fannjiang (ref. 57), not to the authors' own prior results. The benchmark has legitimate fairness and reproducibility limitations--precision settings are not stated, the Dryad link is a placeholder, and phaser's hyperparameters were Optuna-tuned while fold_slice used the reference paper's fixed parameters--but those are confounds in an empirical comparison, not circular reasoning in which an output is equivalent to an input by construction. Accordingly, no circular step meeting the quoted-evidence standard is present.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central software claims rely on standard ptychographic forward models and noise assumptions inherited from the literature, plus a handful of empirically chosen regularization constants and optimizer settings. No new physical entities are introduced.

free parameters (5)
  • Poisson noise epsilon = 0.8 (Si dataset)
    Small offset added to intensities inside logarithms to prevent divergence; chosen empirically and varied in the regularization study (Fig. 7a).
  • object L2 regularization = 0.05
    Penalty on object energy; value chosen by hand or Optuna and varied in the regularization study.
  • object Tikhonov regularization = 0.05
    Penalty on high-frequency object variation; value chosen by hand or Optuna and varied in the regularization study.
  • layers Tikhonov regularization = 5.0 (Si), 10.0 (depth sensitivity)
    Depth-direction smoothing described as critical for multislice quality; values chosen empirically.
  • gradient descent optimizer hyperparameters = not specified in text, provided in Dryad record
    Adam learning rate and schedules tuned via Optuna for quality comparisons; without these values the exact reconstructions cannot be reproduced.
assumptions (4)
  • domain assumption The multislice model with Fresnel propagation is a valid forward model for electron ptychography.
    Invoked throughout Section III; the reconstruction engines and simulations assume this scattering model.
  • domain assumption Detector counts follow the chosen noise statistics (Poisson or variance-stabilized Gaussian) with a small offset epsilon.
    Section III.A defines the likelihood functions; the epsilon offset is rationalized as minimum detector signal.
  • standard math The ptychographic inverse problem is unambiguous under the stated conditions (e.g., known probe positions not on a perfect raster).
    Section III cites Fannjiang [57] to justify recovering probe and object up to a scaling and affine phase ramp.
  • domain assumption Simulated ground-truth objects from Kirkland parameterizations and Debye-Waller thermal averaging are reliable references for error metrics.
    Section VI uses RMS error versus such simulated potentials for hyperoptimization and quality evaluation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of phaser: A unified and extensible framework for fast electron ptychography." pith.science (2026). https://pith.science/paper/5LAIBZUM

@misc{pith2026250514372,
  author       = {Pith},
  title        = {Pith review of: phaser: A unified and extensible framework for fast electron ptychography},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5LAIBZUM}},
  note         = {Machine review of arXiv:2505.14372}
}
read the original abstract

We present \code{phaser}, an open-source Python package that provides a unified interface to both conventional and gradient descent-based ptychographic algorithms. Features such as mixed-state probe, probe position correction, and multislice ptychography make experimental reconstructions practical and robust. Reconstructions are specified in a declarative format and can be run from a command line, Jupyter notebook, or web interface. Multiple computational backends are supported to provide maximum flexibility. With the JAX computational backend, a six-fold improvement in iteration speed is achieved over a widely used package implemented in MATLAB, fold\_slice/PtychoShelves. We report reconstruction success for a variety of experimental datasets, and detail the effects of regularization on convergence and reconstruction quality. The software promises to speed the application and development of ptychographic methods for materials science.

Figures

Figures reproduced from arXiv: 2505.14372 by the authors.

Figure 1
Figure 1. Flowchart of reconstruction engines. Flowcharts describing the (a) conventional and (b) gradient descent engine algorithm. Both engines begin by loading a stored state to the GPU and presimulating the expected diffraction intensity (red blocks). Each reconstruction iteration (black dashed frame) is subdivided into groups of diffraction patterns which are processed in parallel. Several components are implemented thro… view at source ↗
Figure 2
Figure 2. User interface of phaser. a) Example reconstruction plan file for single slice ptychography with the ePIE engine. Reconstruction plans are specified as declarative YAML files. b) Web interface allowing for remote job submission and live viewing of reconstruction process. Probe and object phase are visible as reconstruction proceeds, as well as log messages and errors from the reconstruction. c) Notebook interface, w… view at source ↗
Figure 3
Figure 3. Diagram of server architecture. At the center is a server, which serves as a job queue to schedule jobs given by clients (top) to workers (bottom), located lo￾cally, remotely, or on the cloud. The reconstruction state is returned by workers to the server, which caches the up￾dated data and publishes to client subscribers. Client-server communications take place over HTTP/Websocket transport, while server-worker comm… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: demonstrates the reconstruction performance of phaser. Performance depends heavily on grouping; large groupings are the fastest, at the expense of GPU memory. At small groupings, more time is spent in the relatively-slow Python interpreter. Grouping also affects conver…
Figure 6
Figure 6. Figure 6: fig. 6. Reconstructions were performed with the gradient [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 5
Figure 5. Figure 5: Comparison of reconstruction engines PrScO3 dataset [3] reconstructed with phaser’s ePIE and gra￾dient descent engine, as well as fold slice’s implementation of LSQML. The left column displays reconstructed object mean phase, while the right column displays the frequen…
Figure 6
Figure 6. Figure 6 [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Impact of regularizations on reconstruction. a-d) Convergence of reconstructions while varying reconstruction parameters / regularizers. a) Poisson noise model epsilon, b) layers Tikhonov regularization, c) object L2 regularization and d) object Tikhonov regularization…
Figure 8
Figure 8. Figure 8: Depth sensitivity of reconstructions. Anal￾ysis of reconstruction depth sensitivity for a simulated silicon crystal containing Sn interstitials placed at varying depths. Color represents the reconstructed potential for each intersti￾tial column at each slice in the rec…

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. Guided progressive reconstructive imaging: a new quantization-based framework for low-dose, high-throughput and real-time analytical ptychography

    physics.app-ph 2025-12 conditional novelty 6.0 of 10

    Ptychographic phase reconstruction can be decomposed into a sum of precomputed single-electron contributions, enabling linear-complexity, event-wise direct phase retrieval.

Reference graph

Works this paper leans on

77 extracted references · 76 canonical work pages · cited by 1 Pith paper

  1. [3]

    Z. Chen, Y. Jiang, Y.-T. Shao, et al. , Science 372, 826 (2021)

    work page 2021

  2. [1]

    This update is propagated backwards to the exit plane of the sample

    ePIE In the ePIE algorithm, the noise model is first used to compute a wavefront update χ(⃗k) on the detector. This update is propagated backwards to the exit plane of the sample. Then, at each slice, the wavefront update is split into an update applied to the object slice Oi, and to the previous wavefront/probe Ψi: χn(⃗ r) = F −1(χ(⃗k)) χi−1 = O∗ i (⃗ r)...

    work page

  3. [2]

    LSQML The multislice LSQML algorithm is implemented as de- scribed in Odstrˇ cilet al.[26] and Tsai et al.[21]. Starting with the wavefront update in realspace χn,k(⃗ r), the illu- mination and object update directions are first computed using steepest descent [26, eq 24]: ∆Pi,k(⃗ r) = χi,k(⃗ r)O∗(⃗ r) ∆Oi,k(⃗ r) = χi,k(⃗ r)P ∗(⃗ r) The illumination updat...

    work page

  4. [4]

    Jiang, Z

    Y. Jiang, Z. Chen, Y. Han, et al., Nature 559, 343 (2018)

    work page 2018

  5. [5]

    Z. Chen, M. Odstrcil, Y. Jiang, et al., Nature Communi- cations 11, 2994 (2020)

    work page 2020

  6. [6]

    Chen, Y.-T

    Z. Chen, Y.-T. Shao, Y. Jiang, and D. Muller, Mi- croscopy and Microanalysis 27, 2146 (2021)

    work page 2021

  7. [7]

    H. Sha, J. Cui, and R. Yu, Science Advances 8, eabn2275 (2022)

    work page 2022

  8. [8]

    H. Sha, J. Cui, J. Li, et al., Science Advances 9, eadf1151 (2023)

    work page 2023

Show all 77 references
  1. [9]

    Z. Dong, M. Huo, J. Li, et al. , Nature 630, 847 (2024)

    2024

  2. [10]

    Gilgenbach, X

    C. Gilgenbach, X. Chen, M. Xu, and J. LeBeau, Mi- croscopy and Microanalysis 29, 286 (2023). 13

    2023

  3. [11]

    J. Kim, K. Gordiz, D. Vivona, et al., ACS Nano 18, 31234 (2024)

    2024

  4. [12]

    J. M. Rodenburg, in Advances in Imaging and Electron Physics, Vol. 150, edited by Hawkes (Elsevier, 2008) pp. 87–184

    2008

  5. [13]

    M. Zhu, M. Xu, Y. Yun, et al. , ACS Nano 19, 5568 (2025)

    2025

  6. [14]

    Wirtinger flow

    and ‘enhanced PIE’ (ePIE) [15] was a key break- through, owing to the flexibility and simplicity of such al- gorithms [13]. These algorithms proceed probe position- by-probe position, applying a forward simulation of scat- tering and then updating estimates of the object funct...

    2025 arXiv

  7. [15]

    J. M. Rodenburg and R. H. T. Bates, Philosophical Transactions of the Royal Society of London. Series A: Physical and Engineering Sciences 339, 521 (1992)

    1992

  8. [16]

    Rodenburg and A

    J. Rodenburg and A. Maiden, in Springer Handbook of Microscopy , edited by P. W. Hawkes and J. C. H. Spence (Springer International Publishing, Cham, 2019) pp. 819–904

    2019

  9. [17]

    J. M. Rodenburg and H. M. L. Faulkner, Applied Physics Letters 85, 4795 (2004)

    2004

  10. [18]

    A. M. Maiden and J. M. Rodenburg, Ultramicroscopy 109, 1256 (2009)

    2009

  11. [19]

    J. M. Rodenburg, A. C. Hurst, A. G. Cullis, et al., Phys- ical Review Letters 98, 034801 (2007)

    2007

  12. [20]

    Humphry, B

    M. Humphry, B. Kraus, A. Hurst, et al. , Nature Com- munications 3, 730 (2012)

    2012

  13. [21]

    A. J. D’Alfonso, A. J. Morgan, A. W. C. Yan, et al. , Physical Review B 89, 064101 (2014)

    2014

  14. [22]

    P. Wang, F. Zhang, S. Gao, et al. , Scientific Reports 7, 2857 (2017)

    2017

  15. [23]

    A. M. Maiden, M. J. Humphry, and J. M. Rodenburg, Journal of the Optical Society of America A 29, 1606 (2012)

    2012

  16. [24]

    E. H. R. Tsai, I. Usov, A. Diaz, et al. , Optics Express 24, 29089 (2016)

    2016

  17. [25]

    J. M. Cowley, A. F. Moodie, and IUCr, The scattering of electrons by atoms and crystals. I. A new theoretical approach (1957)

    1957

  18. [26]

    Thibault and A

    P. Thibault and A. Menzel, Nature 494, 68 (2013)

    2013

  19. [27]

    Dwyer, R

    C. Dwyer, R. Erni, and J. Etheridge, Ultramicroscopy Proceedings of the International Workshop on Enhanced Data Generated by Electrons, 110, 952 (2010)

    2010

  20. [28]

    Thibault and M

    P. Thibault and M. Guizar-Sicairos, New Journal of Physics 14, 063004 (2012)

    2012

  21. [29]

    Odstrˇ cil, A

    M. Odstrˇ cil, A. Menzel, and M. Guizar-Sicairos, Optics Express 26, 3108 (2018)

    2018

  22. [30]

    Godard, M

    P. Godard, M. Allain, V. Chamard, and J. Rodenburg, Optics Express 20, 25914 (2012)

    2012

  23. [31]

    Giewekemeyer, P

    K. Giewekemeyer, P. Thibault, S. Kalbfleisch, et al., Pro- ceedings of the National Academy of Sciences 107, 529 (2010)

    2010

  24. [32]

    Thibault, M

    P. Thibault, M. Dierolf, O. Bunk, et al., Ultramicroscopy 109, 338 (2009)

    2009

  25. [33]

    Van den Broek and C

    W. Van den Broek and C. T. Koch, Physical Review B 87, 184108 (2013)

    2013

  26. [34]

    E. J. Cand` es, X. Li, and M. Soltanolkotabi, IEEE Trans- actions on Information Theory 61, 1985 (2015)

    2015

  27. [35]

    Y. S. G. Nashed, T. Peterka, J. Deng, and C. Jacobsen, Procedia Computer Science International Conference on Computational Science, ICCS 2017, 12-14 June 2017, Zurich, Switzerland, 108, 404 (2017)

    2017

  28. [36]

    Kandel, S

    S. Kandel, S. Maddali, M. Allain, et al. , Optics Express 27, 18653 (2019)

    2019

  29. [37]

    Schloz, T

    M. Schloz, T. C. Pekin, Z. Chen, et al. , Optics Express 28, 28306 (2020)

    2020

  30. [38]

    M. Du, S. Kandel, J. Deng, et al. , Optics Express 29, 10000 (2021)

    2021

  31. [39]

    Seifert, D

    J. Seifert, D. Bouchet, L. Loetgering, and A. P. Mosk, OSA Continuum 4, 121 (2021)

    2021

  32. [40]

    Diederichs, Z

    B. Diederichs, Z. Herdegen, A. Strauch, et al. , Nature Communications 15, 101 (2024)

    2024

  33. [41]

    Enders and P

    B. Enders and P. Thibault, Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sci- ences 472, 20160640 (2016)

    2016

  34. [42]

    Wakonig, H.-C

    K. Wakonig, H.-C. Stadler, M. Odstrˇ cil, et al. , Journal of Applied Crystallography 53, 574 (2020)

    2020

  35. [43]

    Loetgering, M

    L. Loetgering, M. Du, D. B. Flaes, et al., Optics Express 31, 13763 (2023)

    2023

  36. [44]

    C.-H. Lee, S. E. Zeltmann, D. Yoon, et al. , PtyRAD: A High-performance and Flexible Ptychographic Re- construction Framework with Automatic Differentiation (2025)

    2025

  37. [45]

    T. B. Edo, D. J. Batey, A. M. Maiden, et al. , Physical Review A 87, 053850 (2013)

    2013

  38. [46]

    Gilgenbach, X

    C. Gilgenbach, X. Chen, and J. M. LeBeau, Microscopy and Microanalysis , ozae055 (2024)

    2024

  39. [47]

    Skoupy, D

    R. Skoupy, D. G. Stroppa, M. Guizar-Sicairos, et al., Mi- croscopy and Microanalysis 30, ozae044.913 (2024)

    2024

  40. [48]

    R. M. Stallman, in Proceedings of the ACM SIGPLAN SIGOA symposium on Text manipulation (Association for Computing Machinery, New York, NY, USA, 1981) pp. 147–156

    1981

  41. [49]

    Froehlich, H

    G. Froehlich, H. J. Hoover, L. Liu, and P. Sorenson, in Proceedings of the 19th international conference on Soft- ware engineering , ICSE ’97 (Association for Computing Machinery, New York, NY, USA, 1997) pp. 491–501

    1997

  42. [50]

    Q. Meng, W. Chen, Y. Wang, et al. , Neurocomputing 337, 46 (2019)

    2019

  43. [51]

    Bradbury, R

    J. Bradbury, R. Frostig, P. Hawkins, et al. , JAX: com- posable transformations of Python+ NumPy programs (2018)

    2018

  44. [52]

    C. R. Harris, K. J. Millman, S. J. van der Walt, et al. , Nature 585, 357 (2020)

    2020

  45. [53]

    Okuta, Y

    R. Okuta, Y. Unno, D. Nishino, et al. , in Proceedings of Workshop on Machine Learning Systems (LearningSys) in The Thirty-first Annual Conference on Neural Infor- mation Processing Systems (NIPS) (2017)

    2017

  46. [54]

    H. Park, J. H. Lee, Y. Oh, et al. , Training Deep Neural Network in Limited Precision (2018)

    2018

  47. [55]

    Kuzmin, M

    A. Kuzmin, M. V. Baalen, Y. Ren, et al., FP8 Quantiza- tion: The Power of the Exponent (2024)

    2024

  48. [56]

    Venkataramani, V

    S. Venkataramani, V. Srinivasan, W. Wang, et al. , in 2021 ACM/IEEE 48th Annual International Symposium on Computer Architecture (ISCA) (2021) pp. 153–166

    2021

  49. [57]

    J. M. LeBeau, S. D. Findlay, L. J. Allen, and S. Stemmer, Ultramicroscopy 110, 118 (2010)

    2010

  50. [58]

    Zhang, Z

    X. Zhang, Z. Chen, Y.-T. Shao, et al. , Microscopy and Microanalysis 31, ozae134 (2025)

    2025

  51. [59]

    E. J. Kirkland, Advanced Computing in Electron Mi- croscopy (Springer US, Boston, MA, 2010)

    2010

  52. [60]

    Fannjiang, Multiscale Modeling & Simulation 17, 973 (2019)

    A. Fannjiang, Multiscale Modeling & Simulation 17, 973 (2019)

    2019

  53. [61]

    M. L. Leidl, B. Diederichs, C. Sachse, and K. M¨ uller- Caspary, Micron 185, 103688 (2024)

    2024

  54. [62]

    Qian and D

    X. Qian and D. Klabjan, The Impact of the Mini-batch Size on the Variance of Gradients in Stochastic Gradient Descent (2020)

    2020

  55. [63]

    Mignacco, P

    F. Mignacco, P. Urbani, and L. Zdeborov´ a, Machine Learning: Science and Technology 2, 035029 (2021)

    2021

  56. [64]

    Babuschkin, K

    DeepMind, I. Babuschkin, K. Baumli, et al. , The Deep- Mind JAX Ecosystem (2020). 14

    2020

  57. [65]

    D. P. Kingma and J. Ba, Adam: A Method for Stochastic Optimization (2017)

    2017

  58. [66]

    Dozat, in ICLR 2016 (San Juan, Puerto Rico, 2016)

    T. Dozat, in ICLR 2016 (San Juan, Puerto Rico, 2016)

    2016

  59. [67]

    Loizou, S

    N. Loizou, S. Vaswani, I. H. Laradji, and S. Lacoste- Julien, in Proceedings of The 24th International Con- ference on Artificial Intelligence and Statistics (PMLR,

  60. [68]

    Tanksalvala, C

    M. Tanksalvala, C. L. Porter, Y. Esashi, et al. , Science Advances 7, eabd9667 (2021)

    2021

  61. [69]

    J. R. Fienup, Applied Optics 21, 2758 (1982)

    1982

  62. [70]

    Elser, JOSA A 20, 40 (2003)

    V. Elser, JOSA A 20, 40 (2003)

    2003

  63. [71]

    M. C. Cao, Z. Chen, Y. Jiang, and Y. Han, Scientific Reports 12, 12284 (2022)

    2022

  64. [72]

    Akiba, S

    T. Akiba, S. Sano, T. Yanase, et al. , Optuna: A Next-generation Hyperparameter Optimization Frame- work (2019)

    2019

  65. [73]

    Bergstra, R

    J. Bergstra, R. Bardenet, Y. Bengio, and B. K´ egl, inAd- vances in Neural Information Processing Systems, Vol. 24 (Curran Associates, Inc., 2011)

    2011

  66. [74]

    Dosselmann and X

    R. Dosselmann and X. D. Yang, Signal, Image and Video Processing 5, 81 (2011)

    2011

  67. [75]

    Nilsson and T

    J. Nilsson and T. Akenine-M¨ oller, Understanding SSIM (2020)

    2020

  68. [76]

    E. J. Verbeke, M. A. Gilles, T. Bendory, and A. Singer, Communications Biology 7, 101 (2024)

    2024

  69. [77]

    T. M. Gesing, R. Uecker, and J.-C. Buhl, Zeitschrift f¨ ur Kristallographie - New Crystal Structures 224, 365 (2009)

    2009

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.