Pith. sign in

REVIEW 4 major objections 5 minor 49 references

Pan-infection Foundation Framework Enables Multiple Pathogen Prediction

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a teacher-student framework built on 35 differential gene pairs can diagnose bacterial and viral infections, specific pathogens, and sepsis from host blood transcriptomes with AUCs up to 0.99.

desk verdict Useful framework, but the evaluation leaks labels through feature selection; the AUCs as reported should not be trusted. read the letter →

arxiv 2501.01462 v1 pith:5IEK33V4 submitted 2024-12-31 cs.LG cs.AIq-bio.GN

Lingrui Zhang , Haonan Wu , Nana Jin , Chenqing Zheng , Jize Xie , Qitai Cai , Jun Wang , Qin Cao
show 3 more authors
Xubin Zheng Jiankun Wang Lixin Cheng
This is my paper · ORCID
classification cs.LGcs.AIq-bio.GN
keywords pan-infectionknowledgedistillationhosttranscriptomegenepairsignaturebloodexpressionpathogenpredictionsepsisdiagnosisteacher-studentmodel
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The authors aim to show that a single pan-infection 'teacher' model, trained on the largest assembled collection of host blood transcriptomes, can transfer knowledge to lightweight 'student' models that diagnose specific pathogens and sepsis. They report that this approach reaches an AUC of 0.97 for distinguishing bacterial from viral infection and AUCs of 0.99, 0.94, 0.93, 0.94, and 0.99 for staphylococcal infection, streptococcal infection, HIV, RSV, and sepsis, respectively. If these numbers hold outside the reported validation sets, the work would offer a deployable, low-parameter diagnostic that could guide antibiotic use and early sepsis care. The authors further claim that the gene-pair signatures across diseases share biological pathways, which they offer as a mechanistic reason for why knowledge transfer works.

What carries the argument

The load-bearing mechanism is PAGE (a pathway-based gene-pair scoring method) combined with Fisher's exact test to select 35 differential gene pairs (DGPs) that represent each disease's host response. These DGPs, computed within biological pathways, transform raw expression values into pairwise comparisons that are more comparable across platforms. A transformer-based teacher model is trained on the pan-infection DGPs, and each student model is trained with a weighted loss combining cross-entropy with a distillation loss that matches the teacher's softened predictions, transferring shared immune-response features.

What would settle it

Rerun the entire TSGPS pipeline with the DGP selection nested inside cross-validation folds (or on training subsets only) and compare the resulting AUCs; if the gap exceeds a few points, the published numbers are inflated by feature-selection leakage and would not generalize to new cohorts.

Watch

Extended reading notes

Core claim

The central discovery is a framework called Teacher-Student Gene Pair Signature (TSGPS) that couples a pan-infection foundation model with distilled pathogen-specific student models. Using 11,247 blood transcriptome samples from 89 datasets, the teacher model learns a general host-response representation that separates health, bacterial infection, and viral infection with an AUC of 0.97. Distilling this teacher into smaller student models yields AUCs of 0.99 for staphylococcal infection, 0.94 for streptococcal infection, 0.93 for HIV, 0.94 for RSV, and 0.99 for sepsis, consistently outperforming vanilla versions of the same student architectures and traditional classifiers. The paper argues that this demonstrates both cross-infection and cross-disease knowledge transfer, and that the compressed students (down to 95.6% parameter reduction) remain accurate enough for clinical deployment.

Load-bearing premise

The 35 differential gene pairs are picked using the entire dataset, including the validation samples, before any train/test split, so the reported accuracy depends on these features being just as informative when chosen only from training data.

Editorial extensions

If this is right

  • If the reported AUCs generalize, a single pan-infection model could support multiple downstream diagnostic tasks without retraining a large model for each pathogen.
  • The 95.6% parameter compression of the extreme student model suggests that host-response diagnostics could run on resource-limited clinical devices.
  • The shared pathways between pan-infection and sepsis models imply that knowledge gained from broad infection data can improve disease-specific screening such as sepsis.
  • The 35-gene-pair signature for each disease is compact enough to be measured with targeted assays, potentially enabling rapid molecular tests.
  • The consistent improvement of distilled students over vanilla models indicates that knowledge distillation is a viable strategy when pathogen-specific training data are scarce.

Reading between the lines

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

  • A natural extension would be to test whether the same teacher-student pipeline transfers to other pathogens with small cohorts, such as fungal or parasitic infections, where the pathway-sharing argument would predict measurable gains.
  • The reported AUCs likely overstate real-world performance because the 35 DGPs were selected using the full dataset before splitting; an honest estimate would require selecting features inside cross-validation folds.
  • If the pathway-based knowledge transfer is real, one could deliberately choose student diseases that share more pathways with the teacher and expect larger distillation gains, a prediction testable on the same data.
  • The sepsis results suggest that this framework might be adapted for other infection-triggered syndromes, such as acute respiratory distress syndrome, though the paper does not explore that direction.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The manuscript proposes TSGPS, a teacher-student knowledge-distillation framework for diagnosing infections from host blood transcriptomes. The authors curate 11,247 samples from GEO, reduce expression profiles to 35 differential gene pairs (DGPs) using their PAGE method, train a pan-infection 'teacher' model (reported AUC 0.97 for both bacterial and viral infection), and distill it into lightweight student models for staphylococcal infection (AUC 0.99), streptococcal infection (0.94), HIV (0.93), RSV (0.94), and sepsis (0.99). The paper also reports parameter compression ratios, comparisons with classical baselines and existing biomarkers, and pathway analyses of the selected gene pairs.

Significance. If the reported numbers held under unbiased evaluation, the paper would provide a valuable lightweight diagnostic pipeline and a large integrated transcriptome resource. The strengths are the scale of the curated compendium, the use of within-sample gene-pair features to mitigate platform effects, the explicit comparison with existing baselines, and the attempt to demonstrate extreme parameter compression. However, the current evaluation does not support the headline claims: feature selection is performed before the train/validation split, no external validation cohorts are used for the pathogen models, and no code or dataset accessions are provided. The central idea is plausible and testable, but the reported AUCs must be re-estimated under a leakage-free protocol before the diagnostic claims can be accepted.

major comments (4)
  1. [Methods, PAGE and Data processing] The 35 DGPs are selected by Fisher's exact test on the full pan-infection and disease-specific datasets before any train/validation split; the text says these DGPs 'are designated as the foundation for both training and validation datasets during model development.' This is selection leakage: the validation samples have influenced which gene pairs are used, so all reported AUCs (0.97 for pan-infection and 0.99/0.94/0.93/0.94/0.99 for the pathogen and sepsis students) are optimistically biased and are not estimates of generalization. Please nest feature selection inside each training fold, or hold out independent cohorts before any feature selection, and re-report all metrics.
  2. [Results, Pan-infection foundation model enhances pathogen prediction] No external validation cohort is used for any of the four pathogen models; the 'validation' is a random split of the same curated data, and the sepsis comparison with SeptiCyte and sNIP uses only four-fold cross-validation. Given the known batch and platform effects in GEO compendia, leave-one-dataset-out or an independent cohort is needed to support the 'generalizable' claim. The manuscript should also report confidence intervals or bootstrap estimates so that the improvements over the vanilla and classical baselines can be assessed.
  3. [Methods, Training TSGPS] The distillation temperature, loss weights, the number of DGPs, and the network architecture are given as fixed values (T=5, weights 0.2 and 0.8, 35 DGPs, number of attention heads and layers), but the manuscript does not state whether these hyperparameters were selected using the validation set. If any model-selection step used the validation data, the reported performance is additionally optimistic; please describe the hyperparameter-selection protocol explicitly or state that no tuning was performed.
  4. [Data availability] The data availability statement provides only a generic GEO URL and refers to 'Fig.2b' for the data used in training and testing; exact GEO accessions, sample labels, and code for PAGE, TSGPS, and the baseline models are not provided. Without these, the curation, the DGP lists, and the reported numbers cannot be reproduced or audited. Please deposit the accession-to-sample mapping and the training/feature-selection code in a public repository.
minor comments (5)
  1. [Abstract and Results] The abstract says 89 blood transcriptome datasets while the Results and Fig.1 say 88; please reconcile the dataset count.
  2. [Methods, PAGE and Data processing] Equation (1) and Table 2 are garbled: the four contingency-table cells a, b, c, and d are not defined, and the displayed formula appears to be a nonstandard version of Fisher's exact test; please define the contingency table and give the standard hypergeometric expression.
  3. [Results, Establishment of pan-infection data] The counts in the text (1,505 bacterial, 5,113 viral, 1,809 sepsis, 2,326 healthy) sum to 10,753, not 11,247, and Table 1's pan-infection row (985 healthy, 6,333 infection, 7,318 total) does not reconcile with the stated totals; please clarify the cohort definitions and overlaps.
  4. [Methods, Training TSGPS] The Methods state that the teacher is trained on pan-infection data 'excluding specific conditions such as sepsis, HIV, RSV, etc.', while Table 1's pan-infection virus count of 3,843 may include HIV and RSV; please clarify exactly which samples are included in the teacher's training set.
  5. [Results, Pan-infection foundation model enhances pathogen prediction] The terms 'original model', 'vanilla model', and 'student model' are used inconsistently (e.g., Fig.5d and the surrounding text); please define each term once and use it consistently throughout.

Circularity Check

1 steps flagged · score 6.0 of 10

Headline AUCs are partially circular: the 35 DGP features are selected with Fisher's exact test on the full dataset before the validation split, so the validation labels already shape the model inputs.

  1. fitted input called prediction [Methods, 'TSGPS / PAGE and Data processing', Eq. 1 and Table 2; Results, 'Pan-infection foundation model']
    "Subsequently, we employ Fisher's exact test to ascertain the p-value statistically, thereby identifying the 35 groups of DGPs that exhibit the most pronounced differential expression according to the following equation 1 and Table 2: ... We procure 35 DGPs about distinct diseases, which are designated as the foundation for both training and validation datasets during model development. ... we reserved 20% of the pan-infection samples as validation set to evaluate its performance."

    The 35 DGP features are selected by applying Fisher's exact test (Eq. 1) to label counts from the entire dataset, including the 20% later 'reserved' as validation. The Methods then states explicitly that these DGPs are 'the foundation for both training and validation datasets.' Consequently, the validation AUC is computed on samples whose labels already contributed to choosing the input features. For high-dimensional gene-pair selection, this guarantees optimistic class separation: the validation set is not independent of the feature-selection step, so the reported AUCs (0.97 pan-infection; 0.99/0.94/0.93/0.94/0.99 for staphylococcus, streptococcus, HIV, RSV, sepsis) are not out-of-sample predictions but re-evaluations of a feature set fit to all labels.

full rationale

The main circular step is the full-data feature selection: the 35 differential gene pairs are chosen by Fisher's exact test on all samples, including the validation split, and then the same features are used to train and validate the models. This makes the headline validation AUCs partially circular in the sense that the validation labels have already influenced the model's input representation. The knowledge-distillation framework itself is not circular: the teacher is trained on pan-infection labels, the students are trained on their own disease labels with a weighted combination of distillation loss and cross-entropy loss, and the reported KD-vs-vanilla improvements are standard supervised comparisons rather than identities forced by definition. Self-citations to PAGE and bvnGPS are method and baseline citations; they are load-bearing for implementation details but do not, by themselves, constitute a circular derivation. The pathway-enrichment characterization is weakened by the fact that gene pairs were constructed inside MSigDB pathways, but the specific shared-pathway findings are empirical rather than strictly forced. Overall, the central performance claim is partially circular because of the pre-split feature selection, though the neural-network training itself is not equivalent to its inputs.

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

No new physical entities are introduced. The model depends on hand-set hyperparameters and the domain assumption that gene-pair features remove batch effects. The most consequential ledger item is the ad hoc use of whole-dataset feature selection, which risks optimistic evaluation.

free parameters (5)
  • Number of differential gene pairs (DGPs) = 35
    Chosen by ranking Fisher exact test p-values across the full datasets; this is the main feature count for every model.
  • Distillation temperature T = 5
    Set in Methods, Training TSGPS; controls the softness of teacher soft targets.
  • Distillation loss weight = 0.2
    Set in Methods; combined with cross-entropy weight 0.8 in the total student loss.
  • Cross-entropy loss weight = 0.8
    Set in Methods; balances ground-truth supervision against distillation.
  • Teacher and student architecture hyperparameters = 2 transformer layers, 5 and 2 attention heads, dropout 0.1, GELU/ReLU activations
    Chosen by hand; reported performance depends on these choices.
assumptions (4)
  • domain assumption Host blood gene expression contains a stable, pathogen-specific signal across 13 countries and 21 platforms.
    The entire diagnostic premise; if host response is not specific enough, no classifier can generalize.
  • domain assumption Intra-pathway gene pair comparisons remove cross-platform and batch effects.
    Inherited from PAGE (ref 7); the paper offers before/after clustering as evidence but no direct batch-effect quantification.
  • ad hoc to paper The 35 DGPs selected with Fisher's exact test on all samples are informative without overfitting.
    Feature selection is done before the reported 80/20 split, so this assumption is load-bearing for the reported AUCs.
  • standard math Standard transformer and AdamW machinery work as specified.
    Background ML machinery, not newly proved in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Pan-infection Foundation Framework Enables Multiple Pathogen Prediction." pith.science (2026). https://pith.science/paper/5IEK33V4

@misc{pith2026250101462,
  author       = {Pith},
  title        = {Pith review of: Pan-infection Foundation Framework Enables Multiple Pathogen Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5IEK33V4}},
  note         = {Machine review of arXiv:2501.01462}
}
read the original abstract

Host-response-based diagnostics can improve the accuracy of diagnosing bacterial and viral infections, thereby reducing inappropriate antibiotic prescriptions. However, the existing cohorts with limited sample size and coarse infections types are unable to support the exploration of an accurate and generalizable diagnostic model. Here, we curate the largest infection host-response transcriptome data, including 11,247 samples across 89 blood transcriptome datasets from 13 countries and 21 platforms. We build a diagnostic model for pathogen prediction starting from a pan-infection model as foundation (AUC = 0.97) based on the pan-infection dataset. Then, we utilize knowledge distillation to efficiently transfer the insights from this "teacher" model to four lightweight pathogen "student" models, i.e., staphylococcal infection (AUC = 0.99), streptococcal infection (AUC = 0.94), HIV infection (AUC = 0.93), and RSV infection (AUC = 0.94), as well as a sepsis "student" model (AUC = 0.99). The proposed knowledge distillation framework not only facilitates the diagnosis of pathogens using pan-infection data, but also enables an across-disease study from pan-infection to sepsis. Moreover, the framework enables high-degree lightweight design of diagnostic models, which is expected to be adaptively deployed in clinical settings.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 37 canonical work pages

  1. [1]

    Kern, W. V. (2010). Management of Staphylococcus aureus bacteremia and endocarditis: progresses and challenges. Current opinion in infectious diseases, 23(4),346-358

    work page 2010

  2. [2]

    J., Jensen, S

    Van Hal, S. J., Jensen, S. O., Vaska, V. L., Espedido,B.A.,Paterson,D.L.,& Gosbell, I. B. (2012). Predictors of mortality in Staphylococcus aureus bacteremia. Clinical microbiologyreviews,25(2),362-386

    work page 2012

  3. [3]

    Wang,F.D.,Chen,Y.Y.,Chen,T.L.,&Liu, C. Y. (2008). Risk factors and mortality in patients with nosocomial Staphylococcus aureus bacteremia. American journal of infectioncontrol,36(2),118-122

    work page 2008

  4. [4]

    Y., Davis, J

    Tong, S. Y., Davis, J. S., Eichenberger, E., Holland, T. L., & Fowler Jr, V. G. (2015). Staphylococcus aureus infections: epidemiology, pathophysiology, clinical manifestations, and management. Clinical microbiologyreviews,28(3),603-661

    work page 2015

  5. [5]

    N., Henningham, A., Gillen, C

    Cole, J. N., Henningham, A., Gillen, C. M., Ramachandran, V., & Walker, M. J. (2008). Human pathogenic streptococcalproteomics and vaccine development. PROTEOMICS–Clinical Applications, 2(3), 387-410

    work page 2008

  6. [6]

    A., Anderson-Berry, A

    Simonsen, K. A., Anderson-Berry, A. L., Delair, S. F., & Davies, H. D. (2014). Early-onset neonatal sepsis. Clinical microbiologyreviews,27(1),21-47

    work page 2014

  7. [7]

    S., Wong, M

    Zheng, X., Leung, K. S., Wong, M. H., & Cheng, L. (2021). Long non-coding RNA pairs to assist in diagnosing sepsis. BMC genomics,22,1-13

    work page 2021

  8. [8]

    M., Townsend, S., Dellinger, R

    Ferrer, R., Martin-Loeches, I., Phillips, G., Osborn, T. M., Townsend, S., Dellinger, R. P., ... & Levy, M. M. (2014). Empiric antibiotic treatment reduces mortality in severe sepsis and septic shock from the first hour: results from a guideline-based performance improvement program. Critical caremedicine,42(8),1749-1755

    work page 2014

Show all 49 references
  1. [9]

    Holcomb,Z.E.,Tsalik,E.L.,Woods,C.W., & McClain, M. T. (2017). Host-based peripheral blood gene expression analysis for diagnosis of infectious diseases. Journal ofClinicalMicrobiology,55(2),360-368

    2017

  2. [10]

    Liang, Y., Zhang, J., Luo, D., Cheng, L., Wang, Y., & Chinese Consortium for the Study of Hepatitis E (CCSHE). (2024). Deregulation of immune response contributing to fulminant hepatitis in HEV infected pregnant women. Journal of MedicalVirology,96(5),e29639

    2024

  3. [11]

    D., Kandukuru, R., Thakur, C., Chakravortty, D.,

    Ravichandran, S., Banerjee, U., DR, G. D., Kandukuru, R., Thakur, C., Chakravortty, D., ... & Chandra, N. (2021). VB10, a new blood biomarker for differential diagnosis and recovery monitoring of acute viral and bacterialinfections.EBioMedicine,67

    2021

  4. [12]

    C., Henao, R., Burke, T

    Lydon, E. C., Henao, R., Burke, T. W., Aydin, M., Nicholson, B. P., Glickman, S. W.,...&Tsalik,E.L.(2019).Validationofa host response test to distinguish bacterial and viral respiratory infection. EBioMedicine,48,453-461

    2019

  5. [13]

    & Geng, Q

    Xie, J., Zheng, X., Yan, J., Li, Q., Jin, N., Wang, S., ... & Geng, Q. (2024). Deep learning model to discriminate diverse infectiontypesbasedonpairwiseanalysisof hostgeneexpression.Iscience,27(6)

    2024

  6. [14]

    Li, Q., Zheng, X., Xie, J.,Wang, R., Li, M., Wong, M. H., ... & Cheng, L. (2023). bvnGPS: a generalizable diagnostic model for acute bacterial and viral infection using integrative host transcriptomics and pretrained neural networks. Bioinformatics, 39(3),btad109

    2023

  7. [15]

    & Liu, X

    Cheng, L., Nan, C., Kang, L., Zhang, N., Liu, S., Chen, H., ... & Liu, X. (2020). Whole blood transcriptomic investigation identifies long non-coding RNAs as regulators in sepsis. Journal of Translational Medicine,18,1-13

    2020

  8. [16]

    T., Rapisarda, A., Sutherland, A

    McHugh, L., Seldon, T.A., Brandon, R.A., Kirk, J. T., Rapisarda, A., Sutherland, A. J., ... & Brandon, R. B. (2015).Amolecular host response assay to discriminate between sepsis and infection-negative systemic inflammation in critically ill patients: discovery and validation i...

    2015

  9. [17]

    P., Klein Klouwenberg, P

    Scicluna, B. P., Klein Klouwenberg, P. M., vanVught,L.A.,Wiewel,M.A.,Ong,D.S., Zwinderman, A. H., ... & van der Poll, T. (2015). A molecular biomarker to diagnose community-acquired pneumonia on intensive care unit admission. American journal of respiratory and critical care m...

    2015

  10. [18]

    E., Perumal, T

    Sweeney, T. E., Perumal, T. M., Henao, R., Nichols,M.,Howrylak,J.A.,Choi,A.M.,... & Langley, R. J. (2018). A community approach to mortality prediction in sepsis via gene expression analysis. Nature communications,9(1),694

    2018

  11. [19]

    H., Leung, K

    Yang,Y.,Zhang,Y.,Li,S.,Zheng,X.,Wong, M. H., Leung, K. S., & Cheng, L. (2021).A robust and generalizable immune-related signature for sepsis diagnostics. IEEE/ACM transactions on computational biology and bioinformatics,19(6),3246-3254

    2021

  12. [20]

    & Cheng, L

    Liu, X., Xu,Y.,Wang, R., Liu, S.,Wang, J., Luo, Y., ... & Cheng, L. (2021). A network-based algorithm for the identification of moonlighting noncoding RNAs and its application in sepsis. Briefingsinbioinformatics,22(1),581-588

    2021

  13. [21]

    Ventola, C. L. (2015). The antibiotic resistance crisis: part 1: causes and threats. Pharmacyandtherapeutics,40(4),277

    2015

  14. [22]

    Wang, J., Horlacher, M., Cheng, L., & Winther, O. (2024). DeepLocRNA: an interpretable deep learning model for predicting RNA subcellular localization with domain-specific transfer-learning. Bioinformatics,40(2),btae065

    2024

  15. [23]

    S., & Cheng, L

    Wang,R.,Zheng,X.,Song,F.,Wong,M.H., Leung, K. S., & Cheng, L. (2023). Deciphering associations between gut microbiota and clinical factors using microbial modules. Bioinformatics, 39(5), btad213

    2023

  16. [24]

    arXiv preprint arXiv:1503.02531

    Hinton,G.(2015).Distilling theKnowledge in a Neural Network. arXiv preprint arXiv:1503.02531

    2015 arXiv

  17. [25]

    E., Chassang, A., Gatta, C., & Bengio, Y

    Romero, A., Ballas, N., Kahou, S. E., Chassang, A., Gatta, C., & Bengio, Y. (2014). Fitnets: Hints for thin deep nets. arXivpreprintarXiv:1412.6550

    2014 arXiv

  18. [26]

    Urban,G.,Geras,K.J.,Kahou,S.E.,Aslan, O.,Wang,S., Caruana, R., ...& Richardson, M. (2016). Do deep convolutional nets really need to be deep and convolutional?. arXivpreprintarXiv:1603.05691

    2016 arXiv

  19. [27]

    Mishra, A., & Marr, D. (2017). Apprentice: Using knowledge distillation techniques to improve low-precision network accuracy. arXiv preprint arXiv:1711.05852

    2017 arXiv

  20. [28]

    Vaswani, A. (2017). Attention is all you need. Advances in Neural Information Processing Systems

    2017

  21. [29]

    & Cheng, L

    Liu, X., Li, N., Liu, S., Wang, J., Zhang, N., Zheng, X., ... & Cheng, L. (2019). Normalization methods for the analysis of unbalanced transcriptome data: a review. Frontiers in bioengineering and biotechnology, 7, 358

    2019

  22. [30]

    E., Li, C., & Rabinovic, A

    Johnson, W. E., Li, C., & Rabinovic, A. (2007). Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics, 8(1), 118-127

    2007

  23. [31]

    P., Wiewel, M

    Scicluna, B. P., Wiewel, M. A., Van Vught, L. A., Hoogendijk, A. J., Klarenbeek, A. M., Franitza, M., ... & Van Der Poll, T. (2018). Molecular biomarker to assist in diagnosing abdominal sepsis upon ICU admission. American journal of respiratory and critical care medicine, 197...

    2018

  24. [32]

    Wang, R., Zheng, X., Wang, J., Wan, S., Song, F., Wong, M. H., ... & Cheng, L. (2022). Improving bulk RNA-seq classification by transferring gene signature from single cells in acute myeloid leukemia. Briefings in Bioinformatics, 23(2), bbac002

    2022

  25. [33]

    & Cheng, L

    Jin, N., Nan, C., Li, W., Lin, P., Xin, Y., Wang, J., ... & Cheng, L. (2025). PAGE-based transfer learning from single-cell to bulk sequencing enhances model generalization for sepsis diagnosis. Briefings in Bioinformatics, 26(1), bbae661

    2025

  26. [34]

    Walawalkar, D., Shen, Z., & Savvides, M. (2020). Online ensemble model compression using knowledge distillation. In Computer Vision – ECCV 2020: 16th European Conference, Glasgow, UK, August 23– 28, 2020, Proceedings, Part XIX 16 (pp. 18-35). Springer International Publishing

    2020

  27. [35]

    Xie, J., Lin, S., Zhang, Y., & Luo, L. (2019). Training convolutional neural networks with cheap convolutions and online distillation. arXiv preprint arXiv:1909.13063

    2019 arXiv

  28. [36]

    Zhang, L., Song, J., Gao, A., Chen, J., Bao, C., & Ma, K. (2019). Be your own teacher: Improve the performance of convolutional neural networks via self distillation. In Proceedings of the IEEE/CVF international conference on computer vision (pp. 3713-3722)

    2019

  29. [37]

    Kim, J., Bhalgat, Y., Lee, J., Patel, C., & Kwak, N. (2019). Qkd: Quantization-aware knowledge distillation. arXiv preprint arXiv:1911.12491

    2019 arXiv

  30. [38]

    Lan, X., Zhu, X., & Gong, S. (2019). Self-referenced deep learning. In Computer Vision–ACCV 2018: 14th Asian Conference on Computer Vision, Perth, Australia, December 2–6, 2018, Revised Selected Papers, Part II 14 (pp. 284-300). Springer International Publishing

    2019

  31. [39]

    Huang, Z., & Wang, N. (2017). Like what you like: Knowledge distill via neuron selectivity transfer. arXiv preprint arXiv:1707.01219

    2017 arXiv

  32. [40]

    Zagoruyko, S., & Komodakis, N. (2016). Paying more attention to attention: Improving the performance of convolutional neural networks via attention transfer. arXiv preprint arXiv:1612.03928

    2016 arXiv

  33. [41]

    (2018, July)

    Wang, H., Zhao, H., Li, X., & Tan, X. (2018, July). Progressive Blockwise Knowledge Distillation for Neural Network Acceleration. In IJCAI (pp. 2769-2775)

    2018

  34. [42]

    Li, T., Li, J., Liu, Z., & Zhang, C. (2020). Few sample knowledge distillation for efficient network compression. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 14639-14647)

    2020

  35. [43]

    Polino, A., Pascanu, R., & Alistarh, D. (2018). Model compression via distillation and quantization. arXiv preprint arXiv:1802.05668

    2018 arXiv

  36. [44]

    Howard, A. G. (2017). Mobilenets: Efficient convolutional neural networks for mobile vision applications. arXiv preprint arXiv:1704.04861

    2017 arXiv

  37. [45]

    Zhang, X., Zhou, X., Lin, M., & Sun, J. (2018). Shufflenet: An extremely efficient convolutional neural network for mobile devices. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 6848-6856)

    2018

  38. [46]

    T., Hovy, E., & Le, Q

    Xie, Q., Luong, M. T., Hovy, E., & Le, Q. V. (2020). Self-training with noisy student improves imagenet classification. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 10687-10698)

    2020

  39. [47]

    Gu, J., & Tresp, V. (2020). Search for better students to learn distilled knowledge. In ECAI 2020 (pp. 1159-1165). IOS Press

    2020

  40. [48]

    Hendrycks, D., & Gimpel, K. (2016). Gaussian error linear units (gelus). arXiv preprint arXiv:1606.08415

    2016 arXiv

  41. [49]

    Nair, V., & Hinton, G. E. (2010). Rectified linear units improve restricted boltzmann machines. In Proceedings of the 27th international conference on machine learning (ICML-10) (pp. 807-814)

    2010

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.