Pith. sign in

REVIEW 4 major objections 4 minor 25 references

MIND: Modality-Informed Knowledge Distillation Framework for Multimodal Clinical Prediction Tasks

T0 review · 4 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read The MIND framework compresses a multimodal clinical network to a third of its size and improves both multimodal and unimodal predictions by distilling weighted ensembles of unimodal teachers into the student.

desk verdict MIND is a well-ablated framework paper with a genuine new combination of known KD ingredients, but the headline gains over the strongest baseline are not statistically pinned down and should be cited with that caveat. read the letter →

arxiv 2502.01158 v1 pith:LFFHTMAJ submitted 2025-02-03 cs.LG cs.AIcs.CV

Alejandro Guerra-Manzanares , Farah E. Shamout This is my paper · ORCID
classification cs.LGcs.AIcs.CV
keywords knowledgedistillationmultimodalfusionmodelcompressionclinicalpredictionensemblechestX-raytimeseriesmodalitybalancing
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 paper claims that a multimodal neural network can be made both smaller and more accurate by training it to imitate an ensemble of larger unimodal teachers, one per input modality, with the imitation weighted so the network can be steered toward the weaker modality. The authors call this MIND and test it on two clinical prediction tasks using chest X-rays plus electronic health record time series, and on three non-medical multimodal benchmarks. On all five tasks, the compressed student outperforms the baselines it is compared with, including the larger MedFuse model it replaces, while also matching or beating unimodal models on single-modality inputs. The paper's point is that distilling unimodal knowledge into per-modality branches of a multimodal fusion network is enough to recover, and exceed, the performance of a much larger fusion network without needing pre-training or modality imputation.

What carries the argument

The load-bearing object is the MIND loss function, which combines three supervised classification losses, one for the fusion head and one for each modality head, with two weighted unimodal ensemble knowledge distillation losses. The distillation loss averages the per-label binary cross-entropy predictions of an ensemble of pre-trained unimodal teachers and asks each student encoder's own head to match them. The two weighting parameters are the instrument that balances modality learning: giving a larger weight to the under-used modality changes the conditional utilization rate and improves that modality's unimodal predictions.

What would settle it

Retrain all baselines using the original MedFuse ResNet-34 image encoder plus the same fused LSTM head, and test whether MIND's 0.782 versus 0.768 AUROC advantage on clinical conditions survives. Alternatively, run a paired statistical test on the 2,161 test-set predictions to see whether the AUROC gap between MIND and the best baseline has a confidence interval that excludes zero.

Watch

Extended reading notes

Core claim

MIND's central discovery is that adding two weighted ensemble-knowledge-distillation losses, one per modality, to a multimodal student's supervised loss transfers the knowledge of pre-trained unimodal teacher ensembles into the student's modality encoders. Each encoder gets its own classification head, so the student can predict from either modality alone at inference. The resulting student is about three times smaller than the original MedFuse architecture in terms of learnable parameters, yet it achieves a higher AUROC on both clinical tasks than every baseline, and its unimodal heads reach or exceed the performance of much larger unimodal models trained on far more data. The framework also gives the user two weights that control how much each modality's distillation matters, which the paper shows can shift the conditional utilization rate and correct imbalance in multimodal learning.

Load-bearing premise

The reported gains over the strongest baseline are real rather than statistical noise, and the baselines truly use the same smaller encoder as the student, so the comparison isolates the distillation method rather than model size.

Editorial extensions

If this is right

  • A single MIND-trained network, roughly one-third the size of the original MedFuse, serves both as a multimodal predictor and as unimodal predictors for each input modality, removing the need to train or deploy separate models for missing-modality cases.
  • The framework extends linearly to more than two modalities, with 2M+1 loss terms for M modalities, so it remains practical as the modality count grows.
  • Because the student is trained from scratch rather than from pre-trained encoders, the gains come from the distillation objective itself, not from transferred weights.
  • Weighting the distillation loss can steer training toward the under-utilized modality and make modality utilization more balanced, which the paper links to improved unimodal encoder performance.

Reading between the lines

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

  • The weighting parameters could likely be scheduled adaptively during training using the conditional utilization rate, rather than fixed as hyperparameters tuned on a validation set, which might reduce tuning cost and further stabilize balance.
  • If the effect is real, MIND may be particularly useful in other small-data multimodal regimes, such as rare-disease imaging combined with genomic or lab data, where unimodal datasets are larger than paired multimodal sets.
  • A natural testable extension is to apply MIND to inference with varying subsets of modalities, since the per-modality heads already provide that capability.
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 / 4 minor

Summary. The paper proposes MIND, a knowledge-distillation framework that trains a compact multimodal student (ResNet-10 plus 2-layer LSTM, 7.5M parameters) from ensembles of larger unimodal teachers. The loss in Equation (6) combines multimodal and unimodal supervision with weighted unimodal ensemble distillation terms. The authors evaluate MIND on multilabel clinical conditions and binary in-hospital mortality prediction using MIMIC-CXR/MIMIC-IV data, and on three non-medical multiclass benchmarks (CREMA-D, S-MNIST, LUMA). They report AUROC/AUPRC gains over MedFuse, MedFuse-3H, TS, MKE, and UME, with bootstrap 95% confidence intervals for the clinical tasks, and an ablation showing a monotonic improvement as components are added.

Significance. If the reported gains are statistically reliable, the paper would make a useful practical contribution: a roughly three-times-smaller multimodal network that improves both fused and unimodal predictions, with a simple, architecture-agnostic loss augmentation and publicly available code. The methodology is clearly described, the multi-head design sensibly handles unimodal samples at inference time, and the ablation study (Table 4) shows a consistent, interpretable trend. The benchmark experiments and the inclusion of both binary and multilabel clinical tasks strengthen the scope. However, the central outperformance claim currently rests on point estimates with overlapping confidence intervals and on a benchmark table without any uncertainty quantification, so the significance of the result is conditional on additional statistical evidence.

major comments (4)
  1. [§5.1, §4, §A.3 (Table 1)] The paper's headline claim that MIND "significantly outperforms" all baselines is not yet supported by the reported statistics. In Table 1, for clinical conditions MIND achieves AUROC 0.782 (95% CI 0.757–0.807) against the strongest baseline TS with 0.768 (0.742–0.794), and for in-hospital mortality 0.844 (0.815–0.872) against 0.828 (0.797–0.857); the bootstrap intervals overlap in both tasks. Moreover, Section A.3 states that each model is the checkpoint with the highest validation AUROC from at least 50 hyperparameter runs, so the reported test numbers are maxima over search and may overstate typical performance. Please report paired per-seed or per-run comparisons (e.g., paired bootstrap of the AUROC differences over the same training runs) and a significance test, or at least the distribution of differences, to substantiate the claimed improvement.
  2. [§4, §A.1, §A.4, Table A1] The description of the baseline architectures is internally inconsistent. Section A.1 states that "we adopt the same configuration for MIND (ResNet-10 and 2-layer LSTM) and all baselines for the sake of comparison", but Section A.4 describes MedFuse as "a ResNet-34 for CXR images and a 2-layer LSTM for time-series data", and Section 4 says the MIND model is three times smaller than MedFuse (Table A1). If the MedFuse and MedFuse-3H rows in Tables 1–4 and B1–B6 used ResNet-34, the comparison confounds the MIND loss with model capacity; if they used ResNet-10, the claimed 3x compression relative to the baseline is inaccurate. Please state explicitly the exact encoder and parameter count used for each baseline row and ensure the size comparison in Section 4 matches the actual experimental configuration.
  3. [§5.5, Table 5] The generalization claims on the three multiclass benchmarks rest on point estimates with no uncertainty quantification. Table 5 reports a single accuracy per model and setting; no confidence intervals, number of seeds, or significance tests are given. For example, on CREMA-D the MIND accuracy is 0.641 vs TS 0.625 and UME 0.620; without an estimate of variance the "significantly outperforms" statement in Section 5.5 is not warranted. Please repeat the benchmark experiments with multiple random seeds (and preferably report mean ± std or bootstrap CIs) or provide paired comparisons.
  4. [§4, §A.2, Tables 2–3] The paper should state explicitly that MIND's student receives knowledge from unimodal teachers trained on substantially larger datasets (124,671 CXR and 42,628 EHR samples) than the paired multimodal training set (7,728 and 4,885 samples), and that this data transfer is part of the method. While the comparison to UME and TS partly accounts for this, the framing in Section 4 that "all MIND and baseline models trained for each specific task are all trained using the same dataset sizes" is potentially misleading when teacher pretraining is considered. Please clarify the exact data available to each model and, if feasible, add a control baseline that receives comparable unimodal pretraining but not the MIND loss.
minor comments (4)
  1. [Algorithm 1] Line 13 of Algorithm 1 ends with "end for=0", which appears to be a typographical error; it should simply be "end for" or "return the trained model".
  2. [Table B6] In Table B6, the MedFuse-3H time-series AUROC entry reads "0.828 0.775, 0.876)"; a parenthesis appears to be missing. Please correct the formatting.
  3. [§5.4] In the utilization-rate equations, the symbol A is used both for modality A and for the accuracy function A(·), which is confusing; consider denoting the accuracy function as Acc(·) or similar.
  4. [Title and Section 1] The capitalization of "Modality-Informed" is inconsistent between the title and the running text (e.g., "Modality-INformed" in the title vs. "modality-informed" in Section 1); please normalize it.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: MIND's objective and evaluation are empirically grounded, with no fitted parameter or self-citation that defines the reported result.

full rationale

This is an empirical framework paper rather than a derivation, and I find no step in which a claimed result is equivalent to its inputs by construction. Equation (6) is a training objective that combines supervised fusion and unimodal losses with weighted ensemble knowledge-distillation terms; each component is defined from teacher predictions, student predictions, and ground-truth labels, not from the test metrics that are later reported. The weighting hyper-parameters omega_A and omega_B are tuned on validation AUROC, as stated in Section 4 and Appendix A.3, and test performance is reported from the checkpoint with the highest validation AUROC; this is standard model selection, not a fitted-input-called-prediction pattern. The ablation study in Section 5.3 compares additive loss components and reports test results, so the contribution of each term is an empirical outcome rather than a definitional consequence. The only noticeable self-citation is the use of MedFuse (Hayat et al., 2022), co-authored by Shamout, as the base architecture and baseline. That citation is not load-bearing for the central claim: MedFuse is the baseline that MIND is compared against and whose encoder is deliberately shrunk from ResNet-34 to ResNet-10, so the cited work is not invoked to establish MIND's superiority. There is no uniqueness theorem imported from the authors' prior work, no ansatz smuggled in via self-citation, and no renaming of a known result as a new organization. The Limitations paragraph appropriately notes the two-task scope and the computational cost of offline distillation, but it asserts no circular dependency. Statistical concerns about overlapping confidence intervals or checkpoint-selected maxima are correctness and reproducibility risks, not circularity, and I do not treat them as evidence of a circular argument.

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

The framework's central result depends on the choice of weighting coefficients omega_A and omega_B, which are tuned per task; these are the main free parameters. The method relies on standard deep learning assumptions and on the correctness of the MIMIC pairing procedure. No new entities are introduced.

free parameters (3)
  • omega_A (CXR weighting) = not fully reported; sensitivity settings 1-500
    Weighting coefficient for CXR distillation loss (Equation 6), tuned per task on validation AUROC; central to the balancing claim.
  • omega_B (EHR weighting) = not fully reported; sensitivity settings 1-500
    Weighting coefficient for EHR distillation loss, tuned per task on validation AUROC.
  • learning rates = ranges 1e-3 to 1e-5 (multimodal), 1e-4 to 1e-6 (unimodal)
    Optimized per model via random search; not fixed constants.
assumptions (3)
  • standard math Standard deep learning toolkit: backpropagation, BCE, softmax, stochastic optimization.
    Unstated background used throughout.
  • domain assumption MIMIC-CXR and MIMIC-IV are correctly linked into paired multimodal samples following Hayat et al.
    Section 4 and A.2 rely on the MedFuse linking procedure; errors here would invalidate the clinical datasets.
  • domain assumption The modified conditional utilization rate formula (with validation AUROC as A) is a valid measure of modality usage in joint fusion.
    Section 5.4 adapts Wu et al. formula to joint fusion; the paper does not validate this diagnostic.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MIND: Modality-Informed Knowledge Distillation Framework for Multimodal Clinical Prediction Tasks." pith.science (2026). https://pith.science/paper/LFFHTMAJ

@misc{pith2026250201158,
  author       = {Pith},
  title        = {Pith review of: MIND: Modality-Informed Knowledge Distillation Framework for Multimodal Clinical Prediction Tasks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LFFHTMAJ}},
  note         = {Machine review of arXiv:2502.01158}
}
read the original abstract

Multimodal fusion leverages information across modalities to learn better feature representations with the goal of improving performance in fusion-based tasks. However, multimodal datasets, especially in medical settings, are typically smaller than their unimodal counterparts, which can impede the performance of multimodal models. Additionally, the increase in the number of modalities is often associated with an overall increase in the size of the multimodal network, which may be undesirable in medical use cases. Utilizing smaller unimodal encoders may lead to sub-optimal performance, particularly when dealing with high-dimensional clinical data. In this paper, we propose the Modality-INformed knowledge Distillation (MIND) framework, a multimodal model compression approach based on knowledge distillation that transfers knowledge from ensembles of pre-trained deep neural networks of varying sizes into a smaller multimodal student. The teacher models consist of unimodal networks, allowing the student to learn from diverse representations. MIND employs multi-head joint fusion models, as opposed to single-head models, enabling the use of unimodal encoders in the case of unimodal samples without requiring imputation or masking of absent modalities. As a result, MIND generates an optimized multimodal model, enhancing both multimodal and unimodal representations. It can also be leveraged to balance multimodal learning during training. We evaluate MIND on binary and multilabel clinical prediction tasks using time series data and chest X-ray images. Additionally, we assess the generalizability of the MIND framework on three non-medical multimodal multiclass datasets. Experimental results demonstrate that MIND enhances the performance of the smaller multimodal network across all five tasks, as well as various fusion methods and multimodal architectures, compared to state-of-the-art baselines.

Figures

Figures reproduced from arXiv: 2502.01158 by the authors.

Figure 1
Figure 1. Architecture of multimodal fusion network. A.1 Architecture for multimodal joint fusion, as in recent work (Hayat et al., 2022). A.2 Modified base architecture that incorporates modality-specific classification heads with the loss shown in Equation 2. B. Architecture of the MIND framework, featuring the loss shown in Equation 5, which integrates unimodal pre-trained teachers for KD into the unimodal encoders. It als… view at source ↗
Figure 2
Figure 2. Label-wise AUROC Performance. Comparison of AUROC performance between the MIND model and the best baseline (TS) for each clinical condition label and group (acute, mixed, and chronic) [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Characterization of modality utilization in multimodal learning for the clinical con￾ditions task. The line graphs depict the conditional utilization rates (ucxr, uehr) and their difference (dutil = ucxr − uehr) every 5 training epochs for the six settings. Vertical lines indicate the epoch with the best AUROC performance for each model on the validation set. as seen in the ωcxr ≫ ωehr setting, which emphasizes the … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 17 canonical work pages

  1. [4]

    Metafed: Federated learning among federations with cyclic knowledge distillation for personalized healthcare.arXiv preprint arXiv:2206.08516,

    Yiqiang Chen, Wang Lu, Xin Qin, Jindong Wang, and Xing Xie. Metafed: Federated learning among federations with cyclic knowledge distillation for personalized healthcare.arXiv preprint arXiv:2206.08516,

    arXiv

  2. [8]

    Medfuse: Multi-modal fusion with clinical time-series data and chest x-ray images.arXiv preprint arXiv:2207.07027,

    Nasir Hayat, Krzysztof J Geras, and Farah E Shamout. Medfuse: Multi-modal fusion with clinical time-series data and chest x-ray images.arXiv preprint arXiv:2207.07027,

    arXiv

  3. [9]

    Distilling the knowledge in a neural network.arXiv preprint arXiv:1503.02531,

    Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the knowledge in a neural network.arXiv preprint arXiv:1503.02531,

    arXiv

  4. [10]

    Knowledge distillation from multi-modal to mono-modal segmentation networks

    Minhao Hu, Matthis Maillard, Ya Zhang, Tommaso Ciceri, Giammarco La Barbera, Isabelle Bloch, and Pietro Gori. Knowledge distillation from multi-modal to mono-modal segmentation networks. InMedical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Lima, Peru, October 4–8, 2020, Proceedings, Part I 23, pp. 772–78...

    2020

  5. [11]

    Vertical federated knowledge transfer via representation distillation for healthcare collaboration networks

    Chung-ju Huang, Leye Wang, and Xiao Han. Vertical federated knowledge transfer via representation distillation for healthcare collaboration networks. InProceedings of the ACM Web Conference 2023, pp. 4188–4199,

    work page 2023

  6. [12]

    Enhancing chinesemulti-labeltextclassificationperformancewithresponse-basedknowledgedistillation

    Szu-Chi Huang, Cheng-Fu Cao, Po-Hsun Liao, Lung-Hao Lee, Po-Lei Lee, and Kuo-Kai Shyu. Enhancing chinesemulti-labeltextclassificationperformancewithresponse-basedknowledgedistillation. In Proceedings of the 34th Conference on Computational Linguistics and Speech Processing (ROCLING 2022), pp. 25–31, 2022a. Yu Huang, Junyang Lin, Chang Zhou, Hongxia Yang, ...

    2022

  7. [14]

    Safenet: A secure, accurate and fast neural network inference

    15 Published in Transactions on Machine Learning Research (01/2025) Qian Lou, Yilin Shen, Hongxia Jin, and Lei Jiang. Safenet: A secure, accurate and fast neural network inference. In International Conference on Learning Representations,

    work page 2025

  8. [16]

    Leukocyte classification using multimodal architecture enhanced by knowledge distillation

    Litao Yang, Deval Mehta, Dwarikanath Mahapatra, and Zongyuan Ge. Leukocyte classification using multimodal architecture enhanced by knowledge distillation. InMedical Optical Imaging and Virtual MicroscopyImageAnalysis: FirstInternationalWorkshop,MOVI2022, HeldinConjunctionwithMICCAI 2022, Singapore, September 18, 2022, Proceedings, pp. 63–72. Springer,

    work page 2022

Show all 25 references
  1. [18]

    For modality A, we use chest X-ray images (CXR) extracted from the MIMIC-CXR dataset (Johnson et al., 2019b)

    23.9 M ResNet-10 and 2-layer LSTM (MIND) 7.5 M A.2 Clinical dataset In our experimental setup, we use two publicly available datasets for two clinical prediction tasks: (i) multilabel clinical conditions prediction and (ii) in-hospital mortality prediction. For modality A, we ...

    2022

  2. [19]

    as the multimodal baseline for our experimental setup. The original implementation of MedFuse employs an LSTM-based fusion module that processes a sequence of modality representations provided by a ResNet-34 encoder for the chest X-ray images and a 2-layer LSTM encoder for the...

    2022

  3. [20]

    It is a typical multimodal architecture proposed for medical tasks, featuring two modality encoders: a ResNet-34 for CXR images and a 2-layer LSTM for time-series data

    serves as the base architecture for all models and experiments in our work. It is a typical multimodal architecture proposed for medical tasks, featuring two modality encoders: a ResNet-34 for CXR images and a 2-layer LSTM for time-series data. The outputs from the encoders ar...

    2020

  4. [21]

    The prediction for a multimodal instance is obtained by weighting the predictions of an ensemble of unimodal models (one model per modality)

    proposes the aggregation of predictions from independently trained unimodal models for multimodal samples on a given task. The prediction for a multimodal instance is obtained by weighting the predictions of an ensemble of unimodal models (one model per modality). Following Du...

    2023

  5. [22]

    B.2 Label-wise AUPRC performance Figure B1 compares the label-wise AUPRC performance of the MIND model with that of TS, the best baseline

    LSA andLSB (independently trained unimodal models) TS (Wang et al., 2020a) LSAB +α×LKDU AB +β×LKDU AB MIND (ours) LSAB +LSA +LSB + ωA×LEKDU A +ωB×LEKDU B A.5 Code for reproducibility To ensure the reproducibility of our results, we make our code available at: https://github.co...

    2025

  6. [23]

    Comparison of AUPRC performance between the MIND model and the best baseline (TS) for each clinical condition label and group (acute, mixed, and chronic)

    0.775 (0.735, 0.813) 0.492 (0.425, 0.570) 0.850 (0.797, 0.900) 0.543 (0.423, 0.679) TS (Wang et al., 2020a) 0.778 (0.738, 0.815) 0.490 (0.423, 0.568) 0.846 (0.795, 0.890) 0.563 (0.449, 0.667) MIND (Ours) 0.790(0.752, 0.826)0.516(0.446, 0.593)0.866(0.817, 0.908)0.572(0.455, 0.6...

    2025

  7. [24]

    It includes 7,442 video clips from 91 actors, each speaking a selection of 12 sentences

    is an audio-visual dataset designed for speech emotion recognition. It includes 7,442 video clips from 91 actors, each speaking a selection of 12 sentences. The utterances express, with 21 Published in Transactions on Machine Learning Research (01/2025) Table B6:Ablation study...

    2025

  8. [25]

    The image modality includes images from a 50-class subset of the CIFAR-10 and CIFAR-100 datasets, while the audio modality contains utterances of the class labels

    is a multimodal dataset designed for benchmarking multimodal learning. The image modality includes images from a 50-class subset of the CIFAR-10 and CIFAR-100 datasets, while the audio modality contains utterances of the class labels. The dataset is imbalanced, with the most p...

    2025

  9. [2012]

    Luma: A benchmark dataset for learning from uncertain and multimodal data.arXiv preprint arXiv:2406.09864,

    Grigor Bezirganyan, Sana Sellami, Laure Berti-Équille, and Sébastien Fournier. Luma: A benchmark dataset for learning from uncertain and multimodal data.arXiv preprint arXiv:2406.09864,

    arXiv

  10. [2014]

    A methodology for training homomorphic encryption friendly neural networks

    Moran Baruch, Nir Drucker, Lev Greenberg, and Guy Moshkowich. A methodology for training homomorphic encryption friendly neural networks. InApplied Cryptography and Network Security Workshops: ACNS 2022 Satellite Workshops, AIBlock, AIHWS, AIoTS, CIMSS, Cloud S&P, SCI, SecMT, ...

    2022

  11. [2018]

    MedFuse (Hayat et al.,

    16 Published in Transactions on Machine Learning Research (01/2025) A Implementation details A.1 Model size Table A1 shows the sizes of the different model architectures. MedFuse (Hayat et al.,

    2025

  12. [2019]

    Multimodal learning with incomplete modalities by knowledge distillation

    Qi Wang, Liang Zhan, Paul Thompson, and Jiayu Zhou. Multimodal learning with incomplete modalities by knowledge distillation. InProceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pp. 1828–1838, 2020a. Weiyao Wang, Du Tran, and Mat...

    arXiv

  13. [2020]

    CRC press,

    14 Published in Transactions on Machine Learning Research (01/2025) Bradley Efron and Robert J Tibshirani.An introduction to the bootstrap. CRC press,

    2025

  14. [2021]

    Ensemble knowledge distillation for learning improved and efficient networks

    Umar Asif, Jianbin Tang, and Stefan Harrer. Ensemble knowledge distillation for learning improved and efficient networks. InECAI 2020, pp. 953–960. IOS Press,

    2020

  15. [2022]

    Enabling multimodal generation on clip via vision-language knowledge distillation.arXiv preprint arXiv:2203.06386,

    Wenliang Dai, Lu Hou, Lifeng Shang, Xin Jiang, Qun Liu, and Pascale Fung. Enabling multimodal generation on clip via vision-language knowledge distillation.arXiv preprint arXiv:2203.06386,

    arXiv

  16. [2023]

    Squeezebiobert: Biobert distillation for healthcare natural language processing

    Hongbin George Du and Yanke Hu. Squeezebiobert: Biobert distillation for healthcare natural language processing. In Computational Data and Social Networks: 9th International Conference, CSoNet 2020, Dallas, TX, USA, December 11–13, 2020, Proceedings 9, pp. 193–201. Springer,

    2020

  17. [2024]

    Alistair EW Johnson, Tom J Pollard, Seth J Berkowitz, Nathaniel R Greenbaum, Matthew P Lungren, Chih-ying Deng, Roger G Mark, and Steven Horng

    URLhttps://doi.org/10.13026/kpb9-mt58. Alistair EW Johnson, Tom J Pollard, Seth J Berkowitz, Nathaniel R Greenbaum, Matthew P Lungren, Chih-ying Deng, Roger G Mark, and Steven Horng. Mimic-cxr, a de-identified publicly available database of chest radiographs with free-text rep...

    doi

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.