Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

Lessons Learned from Deploying Adaptive Machine Learning Agents with Limited Data for Real-time Cell Culture Process Monitoring

T0 review · 3 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Real-time updates with the latest offline measurements keep Raman-based bioreactor models accurate when process conditions shift.

desk verdict A practically useful deployment study whose qualitative lessons are credible, but the quantitative ranking among adaptive methods is statistically thin because each scenario is a single run. read the letter →

arxiv 2509.02606 v1 pith:JFRB7UI4 submitted 2025-08-29 q-bio.QM cs.LG

Thanh Tung Khuat , Johnny Peng , Robert Bassett , Ellen Otte , Bogdan Gabrys This is my paper · ORCID
classification q-bio.QMcs.LG
keywords Ramanspectroscopycellculturemonitoringonlinelearningjust-in-timebioprocesslimiteddatamixtureofexpertsbiopharmaceuticalmanufacturing
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 compares four deployed machine-learning agents that predict glucose, lactate, and ammonium in cell-culture bioreactors from Raman spectra, tested in two industrial case studies with deliberately limited data. The authors aim to establish that adaptive updating with freshly measured offline concentrations is the decisive ingredient: with real-time updates, the just-in-time learning (JITL) agent is consistently accurate, while pretrained models without updates collapse when media, feeding strategy, or cell behaviour shifts. They further show that a simple mixture of experts—the arithmetic average of the four agents—delivers stable, and in the harder case of entirely new media, the best predictions across all three metabolites. The practical value is a direct deployment guideline: choose static pretrained models only for runs close to historical data, and use updating JITL or online agents whenever operating conditions drift.

What carries the argument

The machinery is a deployment loop rather than a single algorithm. JITL rebuilds a local PLSR, PCA+SVR, or KPCA+SVR model for each new Raman query using the 30 most similar samples in a library; the library grows with verified offline measurements from the current run. Online agents (recursive PLSR and online SVR) are pretrained on one run and update incrementally, and the batch models are retrained on the historical library plus new samples; all variants are evaluated with a test-then-train protocol, and a mixture-of-experts solution averages the four predictions. The key work this machinery does is to convert sparse offline samples, collected about twice daily, into rapid adaptation to dis

What would settle it

Repeat each case-study scenario several times under the same nominal conditions and compare the distributions of NMAE and R2 across agents. If JITL with real-time updates is not consistently the most accurate agent across all markers, the paper's central guidance would need to be qualified.

Watch

Extended reading notes

Core claim

The paper's central discovery is that the choice of model matters less than whether the deployment loop includes the latest offline measurements. Under a glucose spike in a familiar process, retrained models trained on 34 historical runs with real-time updates gave the best glucose and lactate predictions, while JITL with an updated data library led on ammonium. Under entirely new feed culture media and feeding strategies, no model without updates generalised; with real-time updates, JITL emerged as the most robust across all markers, and online models trained on a single relevant run often outperformed models retrained on the full historical library because the historical data pulled the mo

Load-bearing premise

The quantitative ranking of the four methods is based on one bioreactor run per scenario, so the differences in NMAE and R2 could reflect quirks of those runs rather than stable properties of the methods.

Editorial extensions

If this is right

  • Deployed Raman monitoring models should receive offline measurements as soon as they are available; scheduled daily updates lag real-time updates.
  • In processes matching historical conditions, retraining on a large historical library plus new data is competitive and often best; in novel conditions, smaller relevant local or online models beat full-library retraining.
  • A simple average of diverse adaptive agents provides smoother and often more accurate real-time metabolite trends, both as a fallback for unknown conditions and as a stabiliser for noisy individual predictions.
  • Model selection guidance can be made conditional on expected process drift: static pretrained models for stable runs, updating JITL or online models for novel media and feeding regimes.

Reading between the lines

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

  • A direct implication the authors leave implicit: the same update-integrating loop should apply to any Raman calibration used for control, not just monitoring, since control decisions inherit the model's drift.
  • The single-run design means the exact error rankings are not transportable; a reasonable next test is repeated or perturbed runs to see whether JITL's robustness holds outside the selected trajectories.
  • The mixture-of-experts result suggests an inexpensive industrial route: retain the individual agents as an audit trail, and use averaging or a simple weighted blend to stabilise predictions when conditions are known to drift.
  • Because offline values arrive only twice daily, pairing adaptive models with automated sampling could reduce the lag between process change and model correction, potentially enabling closed-loop feeding control.
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

3 major / 4 minor

Summary. The paper reports an industrial case-study comparison of four model types—pretrained batch models (PLSR/SVR), online RPLSR and OSVR, just-in-time learning (JITL), and batch retraining—for real-time prediction of glucose, lactate, and ammonium from Raman spectra in fed-batch cell culture. Two case studies are used: one run with a glucose-addition anomaly under conditions similar to historical data, and two runs (one per cell line) using entirely new media and feeding strategies. The central empirical claims are that updating deployed models with new offline measurements is essential, that real-time updating beats daily updating, that JITL and online models generalize better under substantial domain shift than models retrained on the full 34-run library, and that a simple mean ensemble of the four agents improves stability and often accuracy. The evaluation uses held-out runs with a test-then-train protocol and reports NMAE and R².

Significance. If the findings hold, the paper provides practically useful guidance for Raman-based bioprocess monitoring: adaptive updating with verified offline measurements is not optional under process drift, and simple model averaging can mitigate individual-model variability. The use of held-out industrial runs and the test-then-train protocol are appropriate and are real strengths, as is the candid reporting of negative results (e.g., negative R² for non-updated models). However, the quantitative ranking of methods rests on single-run evaluations with no uncertainty bounds, and the selection of the single pretraining run for online models is not disclosed. The qualitative conclusion that updating helps is credible; the quantitative ranking among RPLSR, OSVR, JITL, and retraining is not statistically supported as presented.

major comments (3)
  1. [§3.2, Tables 1–3] The headline ranking among RPLSR, OSVR, JITL, and retraining rests on one bioreactor run per condition class (one run in case study 1, one per cell line in case study 2). Tables 1–3 report single NMAE/R² values with no replicates, confidence intervals, or paired tests. For example, in Table 2 (cell line A), real-time glucose NMAE for OSVR is 5.06 versus 5.46 for JITL; in Table 3 (cell line B) it is 4.99 versus 4.75. Differences of this size cannot be distinguished from run-to-run variability at n=1. The conclusion that JITL is 'the most robust' is an aggregate over six marker-run combinations and is not statistically supported. Please either add repeated runs or uncertainty quantification, or explicitly reframe the contribution as qualitative observations rather than a robust ranking.
  2. [§2.1.1] Online models are pretrained on a single 'carefully selected' bioreactor run, but no selection criteria, representativeness measure, or sensitivity analysis is given. Since the advantage of online/JITL models over retrained models in §3.2 is attributed partly to training on smaller, more relevant data, an uncontrolled pretraining-run choice is a confound: if the selected run happens to be more similar to the test runs than the 34-run library, the benefit may be an artifact of that selection rather than of the adaptive mechanism. Disclose how the pretraining run was chosen and ideally test sensitivity by initializing from an alternative historical run.
  3. [Data availability] The data are proprietary and the figures are normalized to protect intellectual property, so no external check of the quantitative claims is possible. This alone would not be disqualifying for an industrial case study, but combined with single-run evaluation it leaves the central ranking unfalsifiable from the manuscript. Please provide anonymized or perturbed versions of the mapped training/validation/test matrices, or at minimum the test-set predictions for each model so that paired comparisons could be computed.
minor comments (4)
  1. [§3.2] The paragraph beginning 'For lactate predictions, the JITL model using PCA+SVR...' appears twice verbatim. Remove the duplicate.
  2. [§3.2] Typo: 'JILT' should be 'JITL' in the paragraph discussing competitive performance with real-time updating.
  3. [§2.3] The daily-update mechanism is described as triggered at a specific time (e.g., 3:30 PM), but the number of samples accumulated in a 24-hour window is not stated. Please clarify how many training samples are added per daily update.
  4. [§2.4, Eq. (3)] NMAE is normalized by the observed range ymax−ymin; with only ~30 test samples per run, this range is itself a noisy estimate and can be sensitive to a single outlier. Consider reporting an uncertainty interval for NMAE or an alternative robust normalization.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the adaptive-model comparison is a held-out empirical evaluation; the self-citations are methodological provenance, not load-bearing evidence.

full rationale

The paper's central claims—that updating improves predictive accuracy, that JITL with real-time updates is robust under domain shift, and that a simple mixture-of-experts average can improve stability—are empirical findings from two industrial case studies. The case-study runs are held out from training: online models are pretrained on a single historical run, JITL and retrained models use the 34-run library, and hyperparameters are selected via validation or 5-fold cross-validation on historical data, not on the test runs. The test-then-train protocol ensures that updates use labels only after the corresponding prediction is recorded, so the reported NMAE/R2 values are not forced by construction. The self-citations to Khuat et al. (2024a) and Peng et al. (2025b) describe the method provenance (online learning, JITL variants), but the paper independently evaluates these methods on unseen bioreactor runs; no uniqueness theorem is imported, no ansatz is smuggled in as an external constraint, and no fitted parameter is renamed as a prediction. The main weaknesses are statistical (single runs per condition class) and reproducibility (proprietary data), which are external-validity concerns, not circularity. Therefore, no circular step is identified; the score reflects only minor, non-load-bearing self-citation.

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

The central comparison does not hinge on a single magic constant. Hand-set M=30 and preprocessing choices influence all models, while hyperparameters are tuned on validation/historical CV rather than on the test runs. The main unsupported premises are data-quality assumptions and single-run representativeness, not new physical or model constructs.

free parameters (4)
  • JITL neighborhood size M = 30
    Set to 30 in Section 2.1.2; chosen by hand, affects the local model training for every query.
  • Raman spectral window and Savitzky-Golay parameters = 500-3000 cm^-1; window 25, polynomial order 2, first derivative
    Fixed preprocessing choices in Section 2.1.1; different choices would change the input features and similarity metric for all methods.
  • KPCA kernel coefficient and number of components; SVR hyperparameters = Bayesian-optimized over 100 iterations per metabolite
    Tuned on validation data or historical CV (Sections 2.1.1-2.1.3); these fitted values affect the comparison but are not the target results.
  • Number of PLS latent components = Optimized via CV (Optuna, 100 iterations)
    Tuned on the 34-run library for PLSR/RPLSR; standard model selection but data-dependent.
assumptions (4)
  • domain assumption Offline analytical measurements stored in the database are accurate because they are validated by trained operators
    Stated in Section 2.1.1; the entire updating strategy feeds raw offline values into models without noise filtering. If offline values were unreliable, updates would inject noise.
  • ad hoc to paper The single 'carefully selected' pretraining run for online models is representative and its selection is unbiased
    Section 2.1.1 selects one historical run without stating criteria. If the run was chosen with knowledge of the case-study conditions, the online-model advantages are inflated.
  • domain assumption Euclidean distance between preprocessed Raman spectra is a valid similarity metric for retrieving relevant JITL samples under novel media
    Section 2.1.2, Eq. (1)-(2); strong spectral shifts from new media could make nearest-neighbor retrieval chemically misleading.
  • domain assumption A single held-out run and the test-then-train protocol are sufficient to rank the compared methods
    Section 2.4 defines evaluation on one run per condition; no repeated runs, so the ranking lacks uncertainty bounds.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Lessons Learned from Deploying Adaptive Machine Learning Agents with Limited Data for Real-time Cell Culture Process Monitoring." pith.science (2026). https://pith.science/paper/JFRB7UI4

@misc{pith2026250902606,
  author       = {Pith},
  title        = {Pith review of: Lessons Learned from Deploying Adaptive Machine Learning Agents with Limited Data for Real-time Cell Culture Process Monitoring},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JFRB7UI4}},
  note         = {Machine review of arXiv:2509.02606}
}
read the original abstract

This study explores the deployment of three machine learning (ML) approaches for real-time prediction of glucose, lactate, and ammonium concentrations in cell culture processes, using Raman spectroscopy as input features. The research addresses challenges associated with limited data availability and process variability, providing a comparative analysis of pretrained models, just-in-time learning (JITL), and online learning algorithms. Two industrial case studies are presented to evaluate the impact of varying bioprocess conditions on model performance. The findings highlight the specific conditions under which pretrained models demonstrate superior predictive accuracy and identify scenarios where JITL or online learning approaches are more effective for adaptive process monitoring. This study also highlights the critical importance of updating the deployed models/agents with the latest offline analytical measurements during bioreactor operations to maintain the model performance against the changes in cell growth behaviours and operating conditions throughout the bioreactor run. Additionally, the study confirms the usefulness of a simple mixture-of-experts framework in achieving enhanced accuracy and robustness for real-time predictions of metabolite concentrations based on Raman spectral data. These insights contribute to the development of robust strategies for the efficient deployment of ML models in dynamic and changing biomanufacturing environments.

Figures

Figures reproduced from arXiv: 2509.02606 by the authors.

Figure 1
Figure 1. General workflow for online machine learning models from input Raman spectral data and offline analytical metabolites. [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. General workflow of Just-in-time learning. [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. General workflow of pretraining a ML model on all historical data and retraining the model using all historical data and [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: General workflow for the mixture-of-experts approach. [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: The details of offline analytical measurements in case studies. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: The important roles of updating ML models for predictions of [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Performance of all ML models with real-time update for different metabolite measurements in case study [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]
Figure 8
Figure 8. Figure 8: The important roles of updating ML models for predictions of [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]
Figure 9
Figure 9. Figure 9: The important roles of updating ML models for predictions of [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]
Figure 10
Figure 10. Figure 10: Performance of all ML models with real-time update for different metabolite measurements in case study [PITH_FULL_IMAGE:figures/full_fig_p023_10.png]
Figure 11
Figure 11. Figure 11: The outcomes of Mixture-of-experts models and other ML models with real-time update for real-time monitoring of [PITH_FULL_IMAGE:figures/full_fig_p026_11.png]
Figure 12
Figure 12. Figure 12: The outcomes of Mixture-of-experts models and other ML models with real-time update for real-time monitoring of [PITH_FULL_IMAGE:figures/full_fig_p028_12.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Multipath Adaptive Gated Bottleneck Latent ODE with Raman Data Fusion for Cell Culture Process Forecasting

    cs.LG 2026-06 unverdicted novelty 5.5 of 10

    MP-JIT-FT with a gated-bottleneck Latent ODE and Raman fusion ranks best and beats a global Latent ODE on 8/9 targets across 38 heterogeneous fed-batch runs.

Reference graph

Works this paper leans on

33 extracted references · 27 canonical work pages · cited by 1 Pith paper

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  2. [2]

    , author Sano, S

    author Akiba, T. , author Sano, S. , author Yanase, T. , author Ohta, T. , & author Koyama, M. ( year 2019 ). title Optuna: A next-generation hyperparameter optimization framework . In booktitle Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining \/ (pp. pages 2623--2631 )

    work page 2019

  3. [3]

    , author Gavalda, R

    author Bifet, A. , author Gavalda, R. , author Holmes, G. , & author Pfahringer, B. ( year 2023 ). title Machine learning for data streams: with practical examples in MOA \/ . publisher MIT press

    work page 2023

  4. [4]

    author Esmonde-White, K. A. , author Cuellar, M. , & author Lewis, I. R. ( year 2021 ). title The role of raman spectroscopy in biopharmaceuticals from development to manufacturing . journal Analytical and Bioanalytical Chemistry \/ , volume 414 \/ , pages 969--991 . :10.1007/s00216-021-03727-4

    work page doi:10.1007/s00216-021-03727-4 2021

  5. [5]

    , author Umprecht, A

    author Goldrick, S. , author Umprecht, A. , author Tang, A. , author Zakrzewski, R. , author Cheeks, M. , author Turner, R. , author Charles, A. , author Les, K. , author Hulley, M. , author Spencer, C. , & author Farid, S. S. ( year 2020 ). title High-throughput raman spectroscopy combined with innovate data analysis workflow to enhance biopharmaceutical...

    work page doi:10.3390/pr8091179 2020

  6. [6]

    author Grand View Research ( year 2022 ). title Monoclonal antibodies market size, share & trends analysis report by source type (chimeric, murine, humanized, human), by production type (in vivo, in vitro), by application, by end-use, by region, and segment forecasts, 2022-2030 . https://www.grandviewresearch.com/industry-analysis/monoclonal-antibodies-market

    work page 2022

  7. [7]

    , & author Gabrys, B

    author Kadlec, P. , & author Gabrys, B. ( year 2009 ). title Architecture for development of adaptive on-line prediction models . journal Memetic Computing \/ , volume 1 \/ , pages 241--269 . :10.1007/s12293-009-0017-8

    work page doi:10.1007/s12293-009-0017-8 2009

  8. [8]

    , & author Gabrys, B

    author Kadlec, P. , & author Gabrys, B. ( year 2010 ). title Adaptive on-line prediction soft sensing without historical data . In booktitle Proceedings of the 2010 International Joint Conference on Neural Networks ( IJCNN ) \/ (pp. pages 1--8 ). publisher IEEE . :10.1109/ijcnn.2010.5596965

    work page arXiv 2010

Show all 33 references
  1. [9]

    , author Grbi \' c , R

    author Kadlec, P. , author Grbi \' c , R. , & author Gabrys, B. ( year 2011 ). title Review of adaptation mechanisms for data-driven soft sensors . journal Computers & Chemical Engineering \/ , volume 35 \/ , pages 1--24 . :10.1016/j.compchemeng.2010.07.034

    2011 doi

  2. [10]

    author Khuat, T. T. , author Bassett, R. , author Otte, E. , & author Gabrys, B. ( year 2024 a ). title Online machine learning for real-time cell culture process monitoring . In booktitle Proceedings of the 37th Australasian Joint Conference on Artificial Intelligence \/ (pp....

    2024

  3. [11]

    author Khuat, T. T. , author Bassett, R. , author Otte, E. , & author Gabrys, B. ( year 2024 b ). title Uncertainty quantification using ensemble learning and monte carlo sampling for performance prediction and monitoring in cell culture processes . journal arXiv preprint arXi...

    2024 arXiv

  4. [12]

    author Khuat, T. T. , author Bassett, R. , author Otte, E. , author Grevis-James, A. , & author Gabrys, B. ( year 2024 c ). title Applications of machine learning in antibody discovery, process development, manufacturing and formulation: Current trends, challenges, and opportu...

    2024

  5. [13]

    , author Kokol, M

    author Kokol, P. , author Kokol, M. , & author Zagoranski, S. ( year 2022 ). title Machine learning on small size samples: A synthetic knowledge synthesis . journal Science Progress \/ , volume 105 \/ , pages 1--16

    2022

  6. [14]

    , author Theiler, J

    author Ma, J. , author Theiler, J. , & author Perkins, S. ( year 2003 ). title Accurate on-line support vector regression . journal Neural computation \/ , volume 15 \/ , pages 2683--2703

    2003

  7. [15]

    , author Lauri, D

    author Mehdizadeh, H. , author Lauri, D. , author Karry, K. M. , author Moshgbar, M. , author Procopio-Melino, R. , & author Drapeau, D. ( year 2015 ). title Generic raman-based calibration models enabling real-time monitoring of cell culture bioreactors . journal Biotechnolog...

    2015

  8. [16]

    , author Wang, Z

    author Min, R. , author Wang, Z. , author Zhuang, Y. , & author Yi, X. ( year 2023 ). title Application of semi-supervised convolutional neural network regression model based on data augmentation and process spectral labeling in raman predictive modeling of cell culture proces...

    2023

  9. [17]

    author Papathanasiou, M. M. , author Burnak, B. , author Katz, J. , author Shah, N. , & author Pistikopoulos, E. N. ( year 2019 ). title Assisting continuous biomanufacturing through advanced control in downstream purification . journal Computers and Chemical Engineering \/ , ...

    2019 doi

  10. [18]

    , author Park, C.-H

    author Park, S.-Y. , author Park, C.-H. , author Choi, D.-H. , author Hong, J. K. , & author Lee, D.-Y. ( year 2021 ). title Bioprocess digital twins of mammalian cell culture for advanced biomanufacturing . journal Current Opinion in Chemical Engineering \/ , volume 33 \/ , p...

    2021

  11. [19]

    , author Khuat, T

    author Peng, J. , author Khuat, T. T. , author Musial, K. , & author Gabrys, B. ( year 2025 a ). title Machine learning methods for small data and upstream bioprocessing applications: A comprehensive review . journal arXiv preprint arXiv:2506.12322 \/ ,

    2025 arXiv

  12. [20]

    , author Khuat, T

    author Peng, J. , author Khuat, T. T. , author Otte, E. , author Bassett, R. , author Grevis-James, A. , author Musial, K. , & author Gabrys, B. ( year 2025 b ). title Adaptive ensemble-based hyperparameter-free just-in-time learning for robust cell culture process monitoring ...

    2025

  13. [21]

    , author Magill, G

    author Poth, M. , author Magill, G. , author Filgertshofer, A. , author Popp, O. , & author Gro kopf, T. ( year 2022 ). title Extensive evaluation of machine learning models and data preprocessings for raman modeling in bioprocessing . journal Journal of Raman Spectroscopy \/ ...

    2022

  14. [22]

    author Qin, S. J. ( year 1998 ). title Recursive pls algorithms for adaptive data modeling . journal Computers & Chemical Engineering \/ , volume 22 \/ , pages 503--514

    1998

  15. [23]

    , author Khodabandehlou, H

    author Rashedi, M. , author Khodabandehlou, H. , author Wang, T. , author Demers, M. , author Tulsyan, A. , author Garvin, C. , & author Undey, C. ( year 2024 ). title Integration of just-in-time learning with variational autoencoder for cell culture process monitoring based o...

    2024

  16. [24]

    , & author Gabrys, B

    author Ruta, D. , & author Gabrys, B. ( year 2005 ). title Classifier selection for majority voting . journal Information fusion \/ , volume 6 \/ , pages 63--81

    2005

  17. [25]

    , & author Golay, M

    author Savitzky, A. , & author Golay, M. J. ( year 1964 ). title Smoothing and differentiation of data by simplified least squares procedures. journal Analytical chemistry \/ , volume 36 \/ , pages 1627--1639

    1964

  18. [26]

    , author VanAntwerp, J

    author Severson, K. , author VanAntwerp, J. G. , author Natarajan, V. , author Antoniou, C. , author Thömmes, J. , & author Braatz, R. D. ( year 2015 ). title Elastic net with monte carlo sampling for data-based modeling in biopharmaceutical manufacturing facilities . journal ...

    2015 doi

  19. [27]

    , author Lv, G

    author Sun, H. , author Lv, G. , author Mo, J. , author Lv, X. , author Du, G. , & author Liu, Y. ( year 2019 ). title Application of kpca combined with svm in raman spectral discrimination . journal Optik \/ , volume 184 \/ , pages 214--219

    2019

  20. [28]

    , author Khare, S

    author Tulsyan, A. , author Khare, S. , author Huang, B. , author Gopaluni, B. , & author Forbes, F. ( year 2018 ). title A switching strategy for adaptive state estimation . journal Signal Processing \/ , volume 143 \/ , pages 371--380

    2018

  21. [29]

    , author Schorner, G

    author Tulsyan, A. , author Schorner, G. , author Khodabandehlou, H. , author Wang, T. , author Coufal, M. , & author Undey, C. ( year 2019 ). title A machine-learning approach to calibrate generic raman models for real-time monitoring of cell culture processes . journal Biote...

    2019

  22. [30]

    , author Wang, T

    author Tulsyan, A. , author Wang, T. , author Schorner, G. , author Khodabandehlou, H. , author Coufal, M. , & author Undey, C. ( year 2020 ). title Automatic real-time calibration, assessment, and maintenance of generic raman models for online monitoring of cell culture proce...

    2020

  23. [31]

    , author Woon, N

    author Wei, B. , author Woon, N. , author Dai, L. , author Fish, R. , author Tai, M. , author Handagama, W. , author Yin, A. , author Sun, J. , author Maier, A. , author McDaniel, D. , author Kadaub, E. , author Yang, J. , author Saggu, M. , author Woys, A. , author Pester, O....

    2021

  24. [32]

    , author Tulsyan, A

    author Zhao, Z. , author Tulsyan, A. , author Huang, B. , & author Liu, F. ( year 2019 ). title Estimation and identification in batch processes with particle filters . journal Journal of Process Control \/ , volume 81 \/ , pages 1--14

    2019

  25. [33]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.