pith. machine review for the scientific record. sign in

arxiv: 2605.02636 · v1 · submitted 2026-05-04 · 💻 cs.LG · physics.optics

Recognition: 2 theorem links

· Lean Theorem

CNNs for Vis-NIR Chemometrics: From Contradiction to Conditional Design

D\'ario Passos
Authors on Pith no claims yet

Pith reviewed 2026-05-08 18:26 UTC · model grok-4.3

classification 💻 cs.LG physics.optics
keywords CNNVis-NIRchemometricsconditional designreceptive fieldvalidation designspectral physicsnear-infrared spectroscopy
0
0 comments X

The pith

Contradictions in CNN designs for Vis-NIR chemometrics arise from three uncontrolled moderating variables rather than flawed methods.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper examines why CNN studies in near-infrared chemometrics reach opposite conclusions on kernel size, depth, preprocessing, and transfer learning. It traces the inconsistencies to three factors: indirect signals in water-dominated samples, mismatch between network receptive fields and the scale of useful spectral features, and validation choices that function as hidden hyperparameters. The authors replace template-based architecture selection with a conditional framework that ties design decisions to spectral physics, dataset properties, and deployment conditions. This perspective aims to make model comparisons reproducible and aligned with the underlying measurement process.

Core claim

These contradictions are not evidence of irreconcilable methods but a structurally expected consequence of uncontrolled moderating variables: the indirect nature of Vis-NIR measurement in water-dominated matrices, mismatch between effective receptive field and the width of informative spectral structure, and validation design acting as a hidden hyperparameter. Building on evidence from published chemometrics and spectroscopy studies, the paper proposes a conditional design framework that links architecture and preprocessing choices to spectral physics, dataset regime, and intended deployment scenario.

What carries the argument

The conditional design framework that links CNN architecture choices to the physics of indirect Vis-NIR measurements, receptive-field alignment with spectral features, and controlled validation strategies.

If this is right

  • CNN and preprocessing selections must be conditioned on the specific spectral physics and dataset regime instead of universal rules.
  • Validation design, including split strategy and exposure to deployment shifts, must be treated as an explicit hyperparameter to produce reliable model rankings.
  • Practitioners gain reproducible comparisons by controlling the three moderating variables rather than seeking a single best architecture.
  • The field shifts from template-driven selection toward physics-aware and deployment-aligned model design.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same moderating-variable logic could be applied to resolve architecture contradictions in other indirect spectroscopy domains such as Raman or mid-IR.
  • Re-running published datasets with explicit receptive-field matching and standardized validation would provide a direct test of the framework.
  • If the three variables dominate, many reported performance gaps between CNNs and traditional chemometric models may shrink once controls are applied.
  • The perspective suggests treating validation design as a first-class experimental factor in future DL chemometrics benchmarks.

Load-bearing premise

The three listed moderating variables are the dominant and sufficient explanation for the contradictions observed across the literature.

What would settle it

A controlled re-analysis or new multi-study experiment that fixes water-matrix effects, matches receptive fields to feature widths, and standardizes validation splits and tuning budgets, yet still produces inconsistent architecture rankings.

Figures

Figures reproduced from arXiv: 2605.02636 by D\'ario Passos.

Figure 1
Figure 1. Figure 1: Exemple of receptive field (RF) growth in a 1D-CNN with kernel size view at source ↗
read the original abstract

Near-infrared (NIR; a.k.a.\ NIRS) deep-learning studies in chemometrics increasingly report mutually inconsistent conclusions regarding convolutional neural network (CNN) design, including small versus large kernels, shallow versus deep architectures, raw spectra versus preprocessing, and single-domain training versus transfer learning. As a result, the same architecture can appear superior in one study and inferior in another, creating a practical impasse for chemometric practitioners. In this review, we argue that these contradictions are not evidence of irreconcilable methods but a structurally expected consequence of uncontrolled moderating variables. Specifically, we trace recurring disagreements to (i) the indirect nature of Vis--NIR measurement in water-dominated matrices, (ii) mismatch between effective receptive field (ERF) and the width of informative spectral structure, and (iii) validation design (including split strategy, hyperparameter tuning budget, and exposure to deployment-like shifts) acting as a hidden hyperparameter that can dominate model ranking. Building on evidence from published chemometrics and spectroscopy studies, we propose a conditional design framework that links architecture and preprocessing choices to spectral physics, dataset regime, and intended deployment scenario. Overall, the proposed perspective moves DL Chemometrics from template-driven architecture selection toward reproducible, physics-aware, and deployment-aligned model comparison.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript reviews mutually inconsistent conclusions in the Vis-NIR chemometrics literature regarding CNN design choices (small vs. large kernels, shallow vs. deep architectures, raw spectra vs. preprocessing, single-domain vs. transfer learning). It argues that these contradictions arise as a structurally expected consequence of three uncontrolled moderating variables: (i) the indirect nature of Vis-NIR measurements in water-dominated matrices, (ii) mismatch between effective receptive field and the width of informative spectral structure, and (iii) validation design (split strategy, tuning budget, exposure to shifts) acting as a hidden hyperparameter. The authors synthesize evidence from published studies to propose a conditional design framework that links architecture and preprocessing decisions to spectral physics, dataset regime, and deployment scenario, shifting the field from template-driven selection toward reproducible, physics-aware comparisons.

Significance. If the interpretive synthesis holds, the work offers a valuable reframing that could reduce contradictory findings and guide more context-dependent CNN use in chemometrics, emphasizing physics-aware and deployment-aligned evaluation. The absence of new controlled experiments or quantitative meta-analysis limits immediate impact, but the conditional perspective is a constructive alternative to universal design rules and could stimulate targeted follow-up studies.

major comments (2)
  1. [Introduction] Introduction and the section on moderating variables: the central claim that the three listed moderators are the dominant and sufficient explanation for observed contradictions is load-bearing but rests on qualitative selection of studies; the manuscript does not demonstrate (via systematic sampling or frequency analysis) that other factors such as dataset size or noise characteristics are secondary, leaving the sufficiency assertion open to alternative interpretations.
  2. [Conditional Design Framework] The section proposing the conditional design framework: while conceptually linking ERF mismatch and validation design to architecture choices, the framework remains high-level without concrete decision rules, quantitative thresholds (e.g., ERF width relative to spectral feature scale), or worked examples from the cited literature, reducing its actionability for practitioners.
minor comments (2)
  1. Figure captions and the discussion of effective receptive field could more explicitly reference how ERF was estimated or approximated in the referenced CNN studies to allow readers to verify the mismatch argument.
  2. The abstract and introduction use the term 'structurally expected consequence' without a brief formalization or diagram showing how the moderators interact; adding such a schematic would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive report and the recommendation for major revision. The comments highlight opportunities to strengthen the scope and actionability of our synthesis. We address each major comment below and describe the planned revisions.

read point-by-point responses

  1. Referee: [Introduction] Introduction and the section on moderating variables: the central claim that the three listed moderators are the dominant and sufficient explanation for observed contradictions is load-bearing but rests on qualitative selection of studies; the manuscript does not demonstrate (via systematic sampling or frequency analysis) that other factors such as dataset size or noise characteristics are secondary, leaving the sufficiency assertion open to alternative interpretations.

    Authors: We agree that the synthesis is qualitative and does not include systematic sampling or frequency counts across the full literature. The three moderators were selected because they are repeatedly invoked in the cited studies, are directly grounded in the physics of water-dominated Vis-NIR matrices, and map onto standard chemometric validation practices. We do not assert they are the sole or universally dominant factors, only that they provide a coherent explanation for the pattern of contradictions without requiring ad-hoc method critiques. In revision we will add a dedicated paragraph in the moderating-variables section that explicitly discusses alternative factors (dataset size, noise characteristics, instrument variation) and states the rationale for focusing on the three selected moderators. This will clarify the interpretive scope of the review without converting it into a quantitative meta-analysis. revision: partial

  2. Referee: [Conditional Design Framework] The section proposing the conditional design framework: while conceptually linking ERF mismatch and validation design to architecture choices, the framework remains high-level without concrete decision rules, quantitative thresholds (e.g., ERF width relative to spectral feature scale), or worked examples from the cited literature, reducing its actionability for practitioners.

    Authors: We accept that greater concreteness would improve utility. The framework is intentionally conditional rather than prescriptive, but we can illustrate it with specific cases. In the revised manuscript we will expand the framework section with two worked examples drawn from the cited literature (one contrasting small- vs. large-kernel performance under narrow vs. broad spectral features, another showing how validation-split strategy alters apparent superiority of transfer learning). We will also add approximate quantitative guidance, such as relating effective receptive field size to typical widths of water-overtone bands (approximately 10–50 nm) and to the scale of analyte-specific features, based on established NIR spectroscopy references. These additions will make the conditional logic more directly usable while preserving its high-level structure. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The manuscript is a literature review that synthesizes external chemometrics and spectroscopy studies to interpret contradictions in CNN design choices. Its central argument traces inconsistencies to three moderating variables (indirect Vis-NIR measurement, ERF mismatch, and validation design) and proposes a conditional framework linking architecture to spectral physics and deployment. No equations, fitted parameters, or derivations appear that reduce to inputs defined within the paper itself. No self-citations are load-bearing for the core claim, and the synthesis relies on published external evidence rather than internal loops or renamings. The argument remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The review relies on domain assumptions from spectroscopy and machine learning rather than introducing new fitted parameters or invented physical entities.

axioms (2)
  • domain assumption Effective receptive field of a CNN must be matched to the width of chemically informative spectral bands for optimal performance
    Invoked to explain why kernel size recommendations conflict across studies.
  • domain assumption Validation design (split strategy, tuning budget, exposure to shifts) functions as a hidden hyperparameter that can dominate architecture ranking
    Central to the claim that contradictions are structurally expected.

pith-pipeline@v0.9.0 · 5523 in / 1404 out tokens · 32017 ms · 2026-05-08T18:26:24.902547+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

55 extracted references · 49 canonical work pages · 1 internal anchor

  1. [1]

    Tran, Lutgarde M.C

    Jacopo Acquarelli, Twan van Laarhoven, Jan Gerretzen, Thanh N. Tran, Lutgarde M.C. Buydens, and Elena Marchiori. Convolutional neural networks for vibrational spectroscopic data analysis.Analytica Chimica Acta, 954:22–31, February 2017. ISSN 0003-2670. doi: 10.1016/j.aca.2016.12.010. URLhttp://dx.doi.org/10.1016/j.aca.2016.12.010

    work page doi:10.1016/j.aca.2016.12.010 2017

  2. [2]

    Explainable predictive modeling for limited spectral data.Chemometrics and Intelligent Laboratory Systems, 225:104572, June 2022

    Frantishek Akulich, Hadis Anahideh, Manaf Sheyyab, and Dhananjay Ambre. Explainable predictive modeling for limited spectral data.Chemometrics and Intelligent Laboratory Systems, 225:104572, June 2022. ISSN 0169-7439. doi: 10.1016/j.chemolab.2022.104572. URLhttp: //dx.doi.org/10.1016/j.chemolab.2022.104572

    work page doi:10.1016/j.chemolab.2022.104572 2022

  3. [3]

    Review: The evolution of chemometrics coupled with near infrared spectroscopy for fruit quality evaluation.Journal of Near Infrared Spectroscopy, 30(1):3–17, January 2022

    Nicholas T Anderson and Kerry B Walsh. Review: The evolution of chemometrics coupled with near infrared spectroscopy for fruit quality evaluation.Journal of Near Infrared Spectroscopy, 30(1):3–17, January 2022. ISSN 1751-6552. doi: 10.1177/09670335211057235. URLhttp: //dx.doi.org/10.1177/09670335211057235

    work page doi:10.1177/09670335211057235 2022

  4. [4]

    Bec and Christian W

    Krzysztof B. Bec and Christian W. Huck. Breakthrough potential in near-infrared spectroscopy: Spectra simulation. a review of recent developments.Frontiers in Chemistry, 7, February 2019. ISSN 2296-2646. doi: 10.3389/fchem.2019.00048. URLhttp://dx.doi.org/10.3389/fchem. 2019.00048

    work page doi:10.3389/fchem.2019.00048 2019

  5. [5]

    Bec, Justyna Grabska, and Christian W

    Krzysztof B. Bec, Justyna Grabska, and Christian W. Huck. Interpretability in near-infrared (nir) spectroscopy: Current pathways to the long-standing challenge.TrAC Trends in Analytical Chemistry, 189:118254, August 2025. ISSN 0165-9936. doi: 10.1016/j.trac.2025.118254. URL http://dx.doi.org/10.1016/j.trac.2025.118254

    work page doi:10.1016/j.trac.2025.118254 2025

  6. [6]

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, 14 Eric Sigler, Mateusz Lit...

    2020

  7. [7]

    Theory and application of near infrared reflectance spectroscopy in determination of food quality.Trends in Food Science & Technology, 18(2):72–83, February

    Haiyan Cen and Yong He. Theory and application of near infrared reflectance spectroscopy in determination of food quality.Trends in Food Science & Technology, 18(2):72–83, February

  8. [8]

    doi: 10.1016/j.tifs.2006.09.003

    ISSN 0924-2244. doi: 10.1016/j.tifs.2006.09.003. URLhttp://dx.doi.org/10.1016/j. tifs.2006.09.003

    work page doi:10.1016/j.tifs.2006.09.003 2006

  9. [9]

    Endtoend quantitative analysis modeling of nearinfrared spectroscopy based on convolutional neural network.Journal of Chemometrics, 33(5), March

    YuanYuan Chen and ZhiBin Wang. Endtoend quantitative analysis modeling of nearinfrared spectroscopy based on convolutional neural network.Journal of Chemometrics, 33(5), March

  10. [10]

    doi: 10.1002/cem.3122

    ISSN 1099-128X. doi: 10.1002/cem.3122. URLhttp://dx.doi.org/10.1002/cem.3122

    work page doi:10.1002/cem.3122

  11. [11]

    Modern practical convolutional neural networks for multivariate regression: Applications to nir calibration

    Chenhao Cui and Tom Fearn. Modern practical convolutional neural networks for multivariate regression: Applications to nir calibration.Chemometrics and Intelligent Laboratory Systems, 182:9–20, November 2018. ISSN 0169-7439. doi: 10.1016/j.chemolab.2018.07.008. URL http://dx.doi.org/10.1016/j.chemolab.2018.07.008

    work page doi:10.1016/j.chemolab.2018.07.008 2018

  12. [12]

    Automatic neural network hyperparameter optimization for extrapolation: Lessons learned from visible and near-infrared spectroscopy of mango fruit

    Matthew Dirks and David Poole. Automatic neural network hyperparameter optimization for extrapolation: Lessons learned from visible and near-infrared spectroscopy of mango fruit. Chemometrics and Intelligent Laboratory Systems, 231:104685, December 2022. ISSN 0169-7439. doi: 10.1016/j.chemolab.2022.104685. URL http://dx.doi.org/10.1016/j.chemolab.2022. 104685

    work page doi:10.1016/j.chemolab.2022.104685 2022

  13. [13]

    Interpretable perturbator for variable selection in near-infrared spectral analysis.Journal of Chemical Information and Modeling, 64(7):2508–2514, October 2023

    Chaoshu Duan, Xuyang Liu, Wensheng Cai, and Xueguang Shao. Interpretable perturbator for variable selection in near-infrared spectral analysis.Journal of Chemical Information and Modeling, 64(7):2508–2514, October 2023. ISSN 1549-960X. doi: 10.1021/acs.jcim.3c01290. URLhttp://dx.doi.org/10.1021/acs.jcim.3c01290

    work page doi:10.1021/acs.jcim.3c01290 2023

  14. [14]

    Einarson, Andreas Baum, Terkel B

    Kasper A. Einarson, Andreas Baum, Terkel B. Olsen, Jan Larsen, Ibrahim Armagan, Paloma A. Santacoloma, and Line K. H. Clemmensen. Predicting pectin performance strength using nearinfrared spectroscopic data: A comparative evaluation of 1d convolutional neural network, partial least squares, and ridge regression modeling.Journal of Chemometrics, 36(2), May...

    work page doi:10.1002/cem.3348 2021

  15. [15]

    Feng Gan and Jianfei Luo. Simple dilated convolutional neural network for quantitative modeling based on near infrared spectroscopy techniques.Chemometrics and Intelligent Laboratory Systems, 232:104710, January 2023. ISSN 0169-7439. doi: 10.1016/j.chemolab.2022.104710. URLhttp://dx.doi.org/10.1016/j.chemolab.2022.104710

    work page doi:10.1016/j.chemolab.2022.104710 2023

  16. [16]

    Enhancing transferability of near-infrared spectral models for soluble solids content prediction across different fruits.Applied Sciences, 13(9):5417, April 2023

    Cheng Guo, Jin Zhang, Wensheng Cai, and Xueguang Shao. Enhancing transferability of near-infrared spectral models for soluble solids content prediction across different fruits.Applied Sciences, 13(9):5417, April 2023. ISSN 2076-3417. doi: 10.3390/app13095417. URL http: //dx.doi.org/10.3390/app13095417

    work page doi:10.3390/app13095417 2023

  17. [17]

    Haffner, M

    F. Haffner, M. Lacoue-Negre, A. Pirayre, D. Gonçalves, J. Gornay, and M. Moreaud. Ipa: A deep cnn based on inception for petroleum analysis.Fuel, 379:133016, January 2025. ISSN 0016-2361. doi: 10.1016/j.fuel.2024.133016. URLhttp://dx.doi.org/10.1016/j.fuel.2024.133016. 15

    work page doi:10.1016/j.fuel.2024.133016 2025

  18. [18]

    Deep Residual Learning for Image Recognition , isbn =

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778. IEEE, June 2016. doi: 10.1109/cvpr.2016.90. URLhttp://dx.doi.org/10. 1109/CVPR.2016.90

    work page doi:10.1109/cvpr.2016.90 2016

  19. [19]

    On the possible benefits of deep learning for spectral preprocessing.Journal of Chemometrics, 36(2), October

    Runar Helin, Ulf Geir Indahl, Oliver Tomic, and Kristian Hovde Liland. On the possible benefits of deep learning for spectral preprocessing.Journal of Chemometrics, 36(2), October

  20. [20]

    doi: 10.1002/cem.3374

    ISSN 1099-128X. doi: 10.1002/cem.3374. URLhttp://dx.doi.org/10.1002/cem.3374

    work page doi:10.1002/cem.3374

  21. [21]

    Challenges in the use of ai-driven non-destructive spectroscopic tools for rapid food anal- ysis.Foods, 13(6):846, March 2024

    Wenyang Jia, Konstantia Georgouli, Jesus Martinez-Del Rincon, and Anastasios Koidis. Challenges in the use of ai-driven non-destructive spectroscopic tools for rapid food anal- ysis.Foods, 13(6):846, March 2024. ISSN 2304-8158. doi: 10.3390/foods13060846. URL http://dx.doi.org/10.3390/foods13060846

    work page doi:10.3390/foods13060846 2024

  22. [22]

    Kim, Ahyeong Lee, Giyoung Kim, Beom-Soo Shin, and Changyeun Mo

    Min-Jee Kim, Woo-Hyeong Yu, Doo-Jin Song, Seung-Woo Chun, Moon S. Kim, Ahyeong Lee, Giyoung Kim, Beom-Soo Shin, and Changyeun Mo. Prediction of soluble-solid content in citrus fruit using visible–near-infrared hyperspectral imaging based on effective-wavelength selection algorithm.Sensors, 24(5):1512, February 2024. ISSN 1424-8220. doi: 10.3390/s24051512....

    work page doi:10.3390/s24051512 2024

  23. [23]

    Principles and applications of convolutional neural network for spectral analysis in food quality evaluation: A review

    Na Luo, Daming Xu, Bin Xing, Xinting Yang, and Chuanheng Sun. Principles and applications of convolutional neural network for spectral analysis in food quality evaluation: A review. Journal of Food Composition and Analysis, 128:105996, April 2024. ISSN 0889-1575. doi: 10.1016/j.jfca.2024.105996. URLhttp://dx.doi.org/10.1016/j.jfca.2024.105996

    work page doi:10.1016/j.jfca.2024.105996 2024

  24. [24]

    Understanding the effective receptive field in deep convolutional neural networks.arXiv preprint arXiv:1701.04128, 2017

    Wenjie Luo, Yujia Li, Raquel Urtasun, and Richard Zemel. Understanding the effective receptive field in deep convolutional neural networks, 2017. URLhttps://arxiv.org/abs/1701.04128

    work page arXiv 2017

  25. [26]

    Onedimensional convolutional neural networks for spectroscopic signal regression.Journal of Chemometrics, 32(5), November 2017

    Salim Malek, Farid Melgani, and Yakoub Bazi. Onedimensional convolutional neural networks for spectroscopic signal regression.Journal of Chemometrics, 32(5), November 2017. ISSN 1099-128X. doi: 10.1002/cem.2977. URLhttp://dx.doi.org/10.1002/cem.2977

    work page doi:10.1002/cem.2977 2017

  26. [27]

    Martins, R

    J.A. Martins, R. Guerra, R. Pires, M.D. Antunes, T. Panagopoulos, A. Brazio, A.M. Afonso, L. Silva, M.R. Lucas, and A.M. Cavaco. Spectranet–53: A deep residual learning architecture for predicting soluble solids content with vis–nir spectroscopy.Computers and Electronics in Agriculture, 197:106945, June 2022. ISSN 0168-1699. doi: 10.1016/j.compag.2022.106...

    work page doi:10.1016/j.compag.2022.106945 2022

  27. [28]

    Martins, D

    J.A. Martins, D. Rodrigues, A.M. Cavaco, M.D. Antunes, and R. Guerra. Estimation of soluble solids content and fruit temperature in "rocha" pear using vis-nir spectroscopy and the spectranet–32 deep learning architecture.Postharvest Biology and Technology, 199:112281, May 2023. ISSN 0925-5214. doi: 10.1016/j.postharvbio.2023.112281. URLhttp://dx.doi. org/...

    work page doi:10.1016/j.postharvbio.2023.112281 2023

  28. [29]

    Deep calibration transfer: Transferring deep learning models between infrared spectroscopy instruments.Infrared Physics & Technology, 117:103863, 16 September 2021

    Puneet Mishra and Dario Passos. Deep calibration transfer: Transferring deep learning models between infrared spectroscopy instruments.Infrared Physics & Technology, 117:103863, 16 September 2021. ISSN 1350-4495. doi: 10.1016/j.infrared.2021.103863. URLhttp://dx.doi. org/10.1016/j.infrared.2021.103863

    work page doi:10.1016/j.infrared.2021.103863 2021

  29. [30]

    Deep chemometrics: Validation and transfer of a global deep nearinfrared fruit model to use it on a new portable instrument.Journal of Chemometrics, 35(10), July 2021

    Puneet Mishra and Dario Passos. Deep chemometrics: Validation and transfer of a global deep nearinfrared fruit model to use it on a new portable instrument.Journal of Chemometrics, 35(10), July 2021. ISSN 1099-128X. doi: 10.1002/cem.3367. URLhttp://dx.doi.org/10. 1002/cem.3367

    work page doi:10.1002/cem.3367 2021

  30. [31]

    Puneet Mishra and Dario Passos. A synergistic use of chemometrics and deep learning improved the predictive performance of near-infrared spectroscopy models for dry matter prediction in mango fruit.Chemometrics and Intelligent Laboratory Systems, 212:104287, May 2021. ISSN 0169-7439. doi: 10.1016/j.chemolab.2021.104287. URLhttp://dx.doi.org/10.1016/j. che...

    work page doi:10.1016/j.chemolab.2021.104287 2021

  31. [32]

    Fast and power efficient GPU-based explicit elastic wave propagation analysis by low- ordered orthogonal voxel finite element with INT8 tensor cores

    Puneet Mishra and Dario Passos. Multi-output 1-dimensional convolutional neural networks for simultaneous prediction of different traits of fruit based on near-infrared spectroscopy. Postharvest Biology and Technology, 183:111741, January 2022. ISSN 0925-5214. doi: 10.1016/j. postharvbio.2021.111741. URLhttp://dx.doi.org/10.1016/j.postharvbio.2021.111741

    work page doi:10.1016/j 2022

  32. [33]

    Aquaphotomics–from innovative knowledge to inte- grative platform in science and technology.Molecules, 24(15):2742, July 2019

    Jelena Muncan and Roumiana Tsenkova. Aquaphotomics–from innovative knowledge to inte- grative platform in science and technology.Molecules, 24(15):2742, July 2019. ISSN 1420-3049. doi: 10.3390/molecules24152742. URLhttp://dx.doi.org/10.3390/molecules24152742

    work page doi:10.3390/molecules24152742 2019

  33. [34]

    In-context Learning and Induction Heads

    Catherine Olsson, Nelson Elhage, Neel Nanda, Nicholas Joseph, Nova DasSarma, Tom Henighan, Ben Mann, Amanda Askell, Yuntao Bai, Anna Chen, Tom Conerly, Dawn Drain, Deep Ganguli, Zac Hatfield-Dodds, Danny Hernandez, Scott Johnston, Andy Jones, Jackson Kernion, Liane Lovitt, Kamal Ndousse, Dario Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish, a...

    work page internal anchor Pith review doi:10.48550/arxiv.2209.11895 2022

  34. [35]

    Dário Passos. Deep tutti-frutti ii: Explainability of cnn architectures for fruit dry matter predictions.Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 337: 126068, September 2025. ISSN 1386-1425. doi: 10.1016/j.saa.2025.126068. URL http: //dx.doi.org/10.1016/j.saa.2025.126068

    work page doi:10.1016/j.saa.2025.126068 2025

  35. [36]

    Dario Passos and Puneet Mishra. An automated deep learning pipeline based on advanced optimisations for leveraging spectral classification modelling.Chemometrics and Intelligent Laboratory Systems, 215:104354, August 2021. ISSN 0169-7439. doi: 10.1016/j.chemolab.2021. 104354. URLhttp://dx.doi.org/10.1016/j.chemolab.2021.104354

    work page doi:10.1016/j.chemolab.2021 2021

  36. [37]

    Dario Passos and Puneet Mishra. A tutorial on automatic hyperparameter tuning of deep spec- tral modelling for regression and classification tasks.Chemometrics and Intelligent Laboratory Systems, 223:104520, April 2022. ISSN 0169-7439. doi: 10.1016/j.chemolab.2022.104520. URL http://dx.doi.org/10.1016/j.chemolab.2022.104520

    work page doi:10.1016/j.chemolab.2022.104520 2022

  37. [38]

    Dario Passos and Puneet Mishra. Deep tutti frutti: Exploring cnn architectures for dry matter prediction in fruit from multi-fruit near-infrared spectra.Chemometrics and Intelligent Laboratory Systems, 243:105023, December 2023. ISSN 0169-7439. doi: 10.1016/j.chemolab. 2023.105023. URLhttp://dx.doi.org/10.1016/j.chemolab.2023.105023. 17

    work page doi:10.1016/j.chemolab 2023

  38. [39]

    Jean-Michel Roger, Fabien Chauchard, and Veronique Bellon-Maurel. Epo–pls external parameter orthogonalisation of pls application to temperature-independent measurement of sugar content of intact fruits.Chemometrics and Intelligent Laboratory Systems, 66 (2):191–204, June 2003. ISSN 0169-7439. doi: 10.1016/s0169-7439(03)00051-0. URL http://dx.doi.org/10.1...

    work page doi:10.1016/s0169-7439(03)00051-0 2003

  39. [40]

    Preprocessing nir spectra for aquaphotomics.Molecules, 27(20):6795, October 2022

    Jean-Michel Roger, Alexandre Mallet, and Federico Marini. Preprocessing nir spectra for aquaphotomics.Molecules, 27(20):6795, October 2022. ISSN 1420-3049. doi: 10.3390/ molecules27206795. URLhttp://dx.doi.org/10.3390/molecules27206795

    work page doi:10.3390/molecules27206795 2022

  40. [41]

    Independently published, July 2024

    Simone Scardapane.Alice’s Adventures in a Differentiable Wonderland: A Primer on Designing Neural Networks — Volume I: A Tour of the Land. Independently published, July 2024. ISBN 979-8332166181. URL https://www.sscardapane.it/alice-book/. Online version available from the author’s website (CC BY-SA)

    2024

  41. [42]

    Very Deep Convolutional Networks for Large-Scale Image Recognition

    Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image recognition, 2014. URLhttps://arxiv.org/abs/1409.1556

    work page Pith review arXiv 2014

  42. [43]

    Xudong Sun, Phul Subedi, and Kerry B. Walsh. Achieving robustness to temperature change of a nirs-plsr model for intact mango fruit dry matter content.Postharvest Biology and Technology, 162:111117, April 2020. ISSN 0925-5214. doi: 10.1016/j.postharvbio.2019.111117. URLhttp://dx.doi.org/10.1016/j.postharvbio.2019.111117

    work page doi:10.1016/j.postharvbio.2019.111117 2020

  43. [44]

    Matus Telgarsky

    Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Re- thinking the inception architecture for computer vision. In2016 IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 2818–2826. IEEE, June 2016. doi: 10.1109/cvpr.2016.308. URLhttp://dx.doi.org/10.1109/CVPR.2016.308

    work page doi:10.1109/cvpr.2016.308 2016

  44. [45]

    1d-inception-resnet for nir quantitative analysis and its transferability between different spectrometers.Infrared Physics & Technology, 129:104559, March 2023

    Ailing Tan, Yunxin Wang, Yong Zhao, and Yajie Zuo. 1d-inception-resnet for nir quantitative analysis and its transferability between different spectrometers.Infrared Physics & Technology, 129:104559, March 2023. ISSN 1350-4495. doi: 10.1016/j.infrared.2023.104559. URLhttp: //dx.doi.org/10.1016/j.infrared.2023.104559

    work page doi:10.1016/j.infrared.2023.104559 2023

  45. [46]

    Evaluation of 1d convolutional neural network in estimation of mango dry matter content.Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 311:124003, April 2024

    Jeremy Walsh, Arjun Neupane, and Michael Li. Evaluation of 1d convolutional neural network in estimation of mango dry matter content.Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 311:124003, April 2024. ISSN 1386-1425. doi: 10.1016/j.saa.2024. 124003. URLhttp://dx.doi.org/10.1016/j.saa.2024.124003

    work page doi:10.1016/j.saa.2024 2024

  46. [47]

    Walsh, Jose Blasco, Manuela Zude-Sasse, and Xudong Sun

    Kerry B. Walsh, Jose Blasco, Manuela Zude-Sasse, and Xudong Sun. Visible-nir ’point’ spectroscopy in postharvest fruit and vegetable assessment: The science behind three decades of commercial use.Postharvest Biology and Technology, 168:111246, October 2020. ISSN 0925-5214. doi: 10.1016/j.postharvbio.2020.111246. URL http://dx.doi.org/10.1016/j. postharvbi...

    work page doi:10.1016/j.postharvbio.2020.111246 2020

  47. [48]

    Ting, Tao Lin, and Yibin Ying

    Jie Yang, Juntao Li, Jie Hu, Wenjun Yang, Xiaolei Zhang, Jinfan Xu, Youchao Zhang, Xuan Luo, K.C. Ting, Tao Lin, and Yibin Ying. An interpretable deep learning approach for calibration transfer among multiple near-infrared instruments.Computers and Electronics in Agriculture, 192:106584, January 2022. ISSN 0168-1699. doi: 10.1016/j.compag.2021.106584. URL...

    work page doi:10.1016/j.compag.2021.106584 2022

  48. [49]

    Ting, Tao Lin, and Yibin Ying

    Jie Yang, Xuan Luo, Xiaolei Zhang, Dario Passos, Lijuan Xie, Xiuqin Rao, Huirong Xu, K.C. Ting, Tao Lin, and Yibin Ying. A deep learning approach to improving spectral analysis of fruit quality under interseason variation.Food Control, 140:109108, October 2022. ISSN 0956-7135. doi: 10.1016/j.foodcont.2022.109108. URL http://dx.doi.org/10.1016/j.foodcont.2...

    work page doi:10.1016/j.foodcont.2022.109108 2022

  49. [50]

    Multi-Scale Context Aggregation by Dilated Convolutions

    Fisher Yu and Vladlen Koltun. Multi-scale context aggregation by dilated convolutions, 2015. URLhttps://arxiv.org/abs/1511.07122

    work page Pith review arXiv 2015

  50. [51]

    Multiscale deepspectra network: Detection of pyrethroid pesticide residues on the hami melon

    Guowei Yu, Huihui Li, Yujie Li, Yating Hu, Gang Wang, Benxue Ma, and Huting Wang. Multiscale deepspectra network: Detection of pyrethroid pesticide residues on the hami melon. Foods, 12(9):1742, April 2023. ISSN 2304-8158. doi: 10.3390/foods12091742. URL http: //dx.doi.org/10.3390/foods12091742

    work page doi:10.3390/foods12091742 2023

  51. [52]

    A review of machine learning for near-infrared spectroscopy.Sensors, 22(24):9764, December 2022

    Wenwen Zhang, Liyanaarachchi Chamara Kasun, Qi Jie Wang, Yuanjin Zheng, and Zhiping Lin. A review of machine learning for near-infrared spectroscopy.Sensors, 22(24):9764, December 2022. ISSN 1424-8220. doi: 10.3390/s22249764. URL http://dx.doi.org/10. 3390/s22249764

    work page doi:10.3390/s22249764 2022

  52. [53]

    Deepspectra: An end-to-end deep learning approach for quantitative spectral analysis.Analytica Chimica Acta, 1058:48–57, June 2019

    Xiaolei Zhang, Tao Lin, Jinfan Xu, Xuan Luo, and Yibin Ying. Deepspectra: An end-to-end deep learning approach for quantitative spectral analysis.Analytica Chimica Acta, 1058:48–57, June 2019. ISSN 0003-2670. doi: 10.1016/j.aca.2019.01.002. URL http://dx.doi.org/10. 1016/j.aca.2019.01.002

    work page doi:10.1016/j.aca.2019.01.002 2019

  53. [54]

    Understanding the learning mechanism of convolutional neural networks in spectral analysis.Analytica Chimica Acta, 1119:41–51, 2020

    Xiaolei Zhang, Jinfan Xu, Jie Yang, Li Chen, Haibo Zhou, Xiangjiang Liu, Haifeng Li, Tao Lin, and Yibin Ying. Understanding the learning mechanism of convolutional neural networks in spectral analysis.Analytica Chimica Acta, 1119:41–51, 2020. doi: 10.1016/j.aca.2020.03.055. URL https://www.sciencedirect.com/science/article/pii/S0003267020303767. Epub 2020-04-08

    work page doi:10.1016/j.aca.2020.03.055 2020

  54. [55]

    Advanced chemometrics toward robust spectral analysis for fruit quality evaluation.Trends in Food Science & Technology, 150:104612,

    Xiaolei Zhang, Jie Yang, Tao Lin, and Yibin Ying. Advanced chemometrics toward robust spectral analysis for fruit quality evaluation.Trends in Food Science & Technology, 150:104612,

  55. [56]

    URLhttps://doi.org/10.1016/j.tifs.2024.104612

    doi: 10.1016/j.tifs.2024.104612. URLhttps://doi.org/10.1016/j.tifs.2024.104612. 19

    work page doi:10.1016/j.tifs.2024.104612 2024