Pith. sign in

REVIEW 3 major objections 5 minor 40 references

Foundational values for foundation models

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A network of research values explains why medical-imaging researchers adopt foundation models or refuse them.

desk verdict A useful conceptual map of research values for foundation models in medical imaging, but the map's underdetermination makes the 'clear way' claim stronger than the method supports. read the letter →

arxiv 2608.09377 v1 pith:IWLRLFCO submitted 2026-08-10 cs.CY

John S. H. Baxter , Elodie Germani This is my paper · ORCID
classification cs.CY
keywords researchvaluesfoundationmodelsmedicalimagingmachinelearningSocraticmethodologyvaluegraphfairnessreproducibility
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 tries to establish that the choice between using a foundation model, adapting it extensively, or training a model from scratch in medical imaging is systematically driven by a network of research values, not just by performance. Working from a Socratic method, it assembles a value graph in which values such as data efficiency and model reuse support minimal adaptation, while explainability, clinical trust, and knowledge extraction support developing one's own models. The paper's point is that specific research values give researchers coherent, identifiable reasons to adopt or abstain from foundation models. A sympathetic reader would care because this makes a familiar technical disagreement visible as a philosophical one about what makes research good.

What carries the argument

The carrying object is the value graph produced by the Socratic methodology of the authors' prior work [3]. Starting from the technical decision of how much adaptation of a foundation model to perform, the method posits a motivating reason, tests whether the connection is direct or needs an intermediate value, adds that intermediate value, checks for direct conflicts with existing values, and repeats until only fundamental values remain. The resulting directed graph with conflict edges is what lets the paper attribute each technical choice to a chain of instrumental values rather than to a single preference. This machinery works because it converts an abstract question about research culture into concrete, inspectable links between values and the two ends of the adaptation spectrum.

What would settle it

A structured survey or interview study of medical-imaging researchers asking why they adopted or avoided foundation models: if the reasons they give do not match the values and connections in the paper's network, or if a substantial set of researchers cites values the network does not contain, the central claim would be falsified.

Watch

Extended reading notes

Core claim

The central claim is that research values form an instrumental network, presented in the paper's Figure 1, whose structure explains both sides of the foundation-model decision. On the adoption side, data efficiency, model reuse, reproducibility, research speed and facility, model recognition, and research sovereignty each provide a chain of motivation for using a pretrained model with minimal adaptation. On the abstention side, speed and memory consumption, model novelty, explainability, clinical trust, and knowledge extraction motivate extensive adaptation or training from scratch. Some values sit on both sides: environmentalism is split between the cost of creating AI and the cost of using it, publishability inherits its direction from whatever instrumental values it recruits, and fairness currently supports neither side cleanly because foundation models can both broaden representation and perpetuate pretraining biases. The paper concludes that the value network gives the scientific community a more self-reflective understanding of why foundation models spread or fail to spread.

Load-bearing premise

The entire network is the product of the authors' own Socratic reasoning, so if an independent elicitation of medical-imaging researchers' values produced different values or different instrumental links, the claim that these specific values systematically motivate foundation-model use would not be established.

Editorial extensions

If this is right

  • If the network is right, a researcher's choice to adopt or avoid foundation models can be predicted from which values they espouse, and justifications in papers can be read as expressions of these values.
  • The conflict between model novelty and speed and memory consumption shows that values can push in the same direction while opposing each other, so the same technical choice can be overdetermined by different value sets.
  • Fairness does not currently favor either side, so the paper implies that external regulation and auditing, rather than researcher values alone, may be needed to make foundation models fair.
  • Publishability is not a single driver: depending on which instrumental values it recruits, it can justify either using or ignoring foundation models.
  • Future developments such as federated learning could change the value network by adding new benefits and costs, so the map is time-dependent rather than fixed.

Reading between the lines

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

  • One could test the network empirically by content-coding the justification sections of medical-imaging papers that use or reject foundation models and checking whether the stated reasons match the graph's edges.
  • The same Socratic value-graph method could be applied to other contested technical choices, such as adopting federated learning or synthetic data, producing comparable networks with their own ambiguous values.
  • If different research communities espouse different fundamental values, foundation-model adoption should vary systematically across those communities in ways the paper does not try to measure.
  • The graph's claim of completeness is the part most exposed to new evidence, since independent elicitation could add values or redraw edges the authors did not consider.
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 / 5 minor

Summary. The paper explores how "research values"—normatively loaded properties such as data efficiency, explainability, reproducibility, environmentalism, and fairness—can motivate either the use of foundation models or the decision to train/adapt a model from scratch in medical imaging. The authors apply a Socratic methodology (previously introduced in their own work) to construct a graph of fifteen research values and their instrumental/conflict relationships (Figure 1), then discuss each value in turn. Section 5 treats fairness as genuinely ambiguous, and Section 6 concludes that specific research values have "a clear way" of motivating foundation-model use or abstention, contributing to philosophy of machine learning in medicine. The paper is explicitly exploratory and relies on introspective reasoning supplemented by selected citations rather than on empirical elicitation or a systematic literature review.

Significance. If the central conclusion is accepted in a suitably qualified form, the paper makes a worthwhile conceptual contribution: it gives a structured, discussable map of why researchers in medical image analysis might legitimately justify either using foundation models or abstaining from them. The individual value discussions are generally explicit and many are supported by relevant citations, and the paper is honest about its exploratory nature and about exceptions such as environmentalism, publishability, and fairness. The main value of the contribution is therefore as a hypothesis-generating philosophical taxonomy, not as an empirical measurement of how researchers actually reason. Its broader significance depends on whether the graph is presented as a plausible possible map rather than as a uniquely determined or complete one; that calibration is the main point that needs revision.

major comments (3)
  1. [Section 6] The central claim that "specific research values do have a clear way of motivating the use of foundation models or abstaining from their use" is stronger than the paper's own evidence supports. Sections 4 and 5 explicitly state that environmentalism "is torn" between creation and use, that publishability "should not directly motivate a specific technical decision in a consistent way," and that fairness "could be argued to motivate either side." With at least three of the fifteen values not having a clear side assignment, the concluding sentence should be qualified, e.g., to "most of the values examined here" or "values often have a clear way, with notable exceptions such as fairness."
  2. [Section 3 and Section 4] The construction of the graph is underdetermined by the stated methodology. Step 5 of the Socratic procedure says to repeat until all values are "fundamental or have already been explored," and Section 4 says the process stopped when "we had exhausted all the values that could be cleanly associated" with one side. This stopping rule and the directionality of edges (e.g., why model recognition is instrumental for publishability rather than vice versa) rest on the authors' introspection, with no operational criterion or inter-rater check. Because the central claim relies on Figure 1 as a faithful representation, the paper should either provide a more explicit coding protocol and reliability evidence, present the graph as one possible reconstruction, or weaken the conclusion accordingly.
  3. [Figure 1 and Section 4, Model novelty] The meaning of the conflict edge is unclear for the model-novelty versus speed/memory relationship. The text states that "despite this contradiction, there does not appear to be a possibility of an internal tension on the part of the researcher" and that both values "seem to both support the right side of the spectrum." Yet the legend defines a conflict as "A and B are in conflict," and the figure draws a conflict edge between these two values. The paper should clarify whether conflict edges represent opposition in instrumental reasoning, opposition in side assignment, or something else; as written, the graph and text are hard to reconcile.
minor comments (5)
  1. [Figure 1 legend] The legend entry "A is potentially an instrumental value for A, i.e. A could motivate B" appears to contain a typographical error and a possibly reversed relation; it should read "B is potentially an instrumental value for A" or the sentence should be rephrased so the arrow direction is unambiguous.
  2. [Section 4, Clinical trust] The sentence contains "explainations," which appears to be a typo for "explanations."
  3. [Section 4, Reproducibility] The phrase "reproducibility is complicated highly from the presence" should read "is highly complicated by the presence."
  4. [Section 4, Model recognition] The text says "modal recognition" in the first sentence; this should be "model recognition" to match the section title and the graph.
  5. [Section 4, Data efficiency] The phrase "researchers need fewer datasets to train a particular model" mixes count and mass nouns; it should be "less data" or "fewer training examples" for consistency.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity; the value network is a new reasoned construction, and the self-citation to the authors' Socratic method is not load-bearing.

full rationale

The paper's derivation chain is not circular. It defines research values (Section 3), restates the Socratic methodology from Baxter & Eagleson [3] in its own words, applies it to the foundation-model decision (Section 4), and concludes that specific values have a clear way of motivating use or abstention (Section 6). The values in Figure 1 are not fitted parameters and no prediction is statistically forced: each edge is a reasoned instrumental claim, several of which are independently cited (e.g., data efficiency [37], model reuse [13], explainability [2,20,30]). The self-citation to [3] is used for the method and for the continuous-spectrum framing, but the paper gives an independent argument for the spectrum in Section 2 ('as the adaptation becomes more complex, less of the underlying foundation model remains'), and the method itself is fully described in Section 3, so the citation is not load-bearing. The one quasi-tautological aspect is that the Socratic procedure selects values as hypothesized reasons for or against the technical decision, so the concluding summary is partly a restatement of the graph; however, the graph's specific content (e.g., the model-novelty versus speed/memory tension, the instrumentality of model recognition) is not implied by the method alone and could be wrong or incomplete. The paper explicitly acknowledges this dependence in Section 6 ('the results may be limited by the scope of experience for the particular authors'), and Section 5's treatment of fairness as ambiguous shows that the method does not automatically force every value onto one side. Underdetermination by the authors' introspection is a validity limitation, not a circular reduction, and no equation or fitted quantity is renamed as a prediction. Minor self-citation justifies a score of 1 rather than 0.

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

No free parameters or invented entities; the paper is conceptual. The main assumptions are the validity of the self-cited Socratic method and the completeness of the value set chosen by the authors.

assumptions (3)
  • domain assumption The Socratic method from Baxter & Eagleson [3] is a valid way to map research values for foundation models in medical imaging.
    The paper adopts the method from a prior paper by the same authors without independent validation, and the graph's validity rests on this method.
  • ad hoc to paper The set of values considered (data efficiency, model reuse, speed, memory, novelty, explainability, reproducibility, research speed, clinical trust, knowledge extraction, model recognition, research sovereignty, environmentalism, publishability, fairness) is sufficiently complete for the analysis.
    The values are selected based on the authors' experience; the paper acknowledges this limitation in the Discussion.
  • domain assumption The cited empirical findings about foundation models (e.g., pretraining on single datasets, fairness issues) are accurate.
    The arguments rely on these citations to support the edges in the graph.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Foundational values for foundation models." pith.science (2026). https://pith.science/paper/IWLRLFCO

@misc{pith2026260809377,
  author       = {Pith},
  title        = {Pith review of: Foundational values for foundation models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IWLRLFCO}},
  note         = {Machine review of arXiv:2608.09377}
}
read the original abstract

Research values, properties with a distinctive normative dimension, often affect how technological research is performed in both direct and indirect ways by influencing how technical decisions are made. In machine learning for medical imaging, understanding these values can be important for understanding why particular researchers justify the decisions made in their publications and explain why certain technologies become ubiquitous (or not) in the scientific literature and in the clinic. This article explores one of these technologies, foundation models, finding detailed justifications both for their use and abstention from their use. By taking a Socratic approach to research values arising from this specific technical decision, this article aims to better illustrate how foundation models fit into the philosophy of machine learning in medicine.

Figures

Figures reproduced from arXiv: 2608.09377 by the authors.

Figure 1
Figure 1. Graph of research values related to the degree of foundation model use. Data efficiency The primary stated motivation for the use of foundation models is data efficiency [37] (or equivalently, the ability to overcome data scarcity), which means that researchers need fewer datasets to train a particular model, especially in the case where zero-shot or similar methodologies are used and there are few added components … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 29 canonical work pages

  1. [3]

    Medical Image Analysis102, 103494 (2025)

    Baxter, J.S., Eagleson, R.: Exploring the values underlying machine learning re- search in medical image analysis. Medical Image Analysis102, 103494 (2025)

    work page 2025

  2. [1]

    Statistical Science41(2), 316–353 (2026)

    Arbel, J., Pitas, K., Vladimirova, M., Fortuin, V.: A primer on bayesian neural networks: review and debates. Statistical Science41(2), 316–353 (2026)

    work page 2026

  3. [2]

    arXiv preprint arXiv:2509.24231 (2025)

    Bai, Y., Cheng, H., Zhou, Y., Zhou, J., Thirunavukarasu, A., Ke, Y., Yao, J., Fukutsu, K., Quek, C.W.N., Hong, A., et al.: Evlf-fm: Explainable vision language foundation model for medicine. arXiv preprint arXiv:2509.24231 (2025)

    work page arXiv 2025

  4. [4]

    Research Square pp

    Blankemeier, L., Cohen, J.P., Kumar, A., Van Veen, D., Gardezi, S.J.S., Paschali, M., Chen, Z., Delbrouck, J.B., Reis, E., Truyts, C., et al.: Merlin: A vision language foundation model for 3d computed tomography. Research Square pp. rs–3 (2024)

    2024

  5. [5]

    Information Systems p

    Curcio, E.: Evaluating the lifecycle economics of ai: The levelized cost of artificial intelligence (lcoai). Information Systems p. 102634 (2025)

    work page 2025

  6. [6]

    Counsel- ing and values53(1), 22–38 (2009) Foundational values for foundation models 11

    Duffy, M., Chenail, R.J.: Values in qualitative and quantitative research. Counsel- ing and values53(1), 22–38 (2009) Foundational values for foundation models 11

    work page 2009

  7. [7]

    Human Brain Map- ping43(16), 4835–4851 (2022)

    Estudillo-Romero, A., Haegelen, C., Jannin, P., Baxter, J.S.: Voxel-based diktiom- etry: Combining convolutional neural networks with voxel-based analysis and its application in diffusion tensor imaging for parkinson’s disease. Human Brain Map- ping43(16), 4835–4851 (2022)

    work page 2022

  8. [8]

    Neu- roimage: Reports4(2), 100202 (2024)

    Estudillo-Romero, A., Migliaccio, R., Batrancourt, B., Jannin, P., Baxter, J.S.: Non-local diffusion-based biomarkers in patients with cocaine use disorder. Neu- roimage: Reports4(2), 100202 (2024)

    work page 2024

Show all 40 references
  1. [9]

    Partners Universal International Innova- tion Journal1(2), 97–104 (2023)

    George, A.S., George, A.H., Martin, A.G.: The environmental impact of ai: a case study of water consumption by chat gpt. Partners Universal International Innova- tion Journal1(2), 97–104 (2023)

    2023

  2. [10]

    Soft Computing29(13), 4823–4856 (2025)

    Girdhar, N., Raj, A., Sharma, D., Singh, V., Doucet, A., Renz, M.: A comprehen- sive review of frugal artificial intelligence: challenges, applications, and the road to sustainable ai. Soft Computing29(13), 4823–4856 (2025)

    2025

  3. [11]

    Advances in neural information processing systems32(2019)

    Ignatiev, A., Narodytska, N., Marques-Silva, J.: On relating explanations and ad- versarial examples. Advances in neural information processing systems32(2019)

    2019

  4. [12]

    IEEE Transactions on Artificial Intelligence 4(4), 959–971 (2022)

    Jiang, Z., Wang, Y., Li, C.T., Angelov, P., Jiang, R.: Delve into neural activations: Toward understanding dying neurons. IEEE Transactions on Artificial Intelligence 4(4), 959–971 (2022)

    2022

  5. [13]

    IEEE Reviews in Biomedical Engineering (2025)

    Khan, W., Leem, S., See, K.B., Wong, J.K., Zhang, S., Fang, R.: A comprehensive survey of foundation models in medicine. IEEE Reviews in Biomedical Engineering (2025)

    2025

  6. [14]

    arXiv preprint arXiv:2604.26482 (2026)

    Klotz, D.: The buy-or-build decision, revisited: How agentic ai changes the eco- nomics of enterprise software. arXiv preprint arXiv:2604.26482 (2026)

    2026 arXiv

  7. [15]

    CAAI Transactions on Intelligence Technology9(2), 286–302 (2024)

    Lin, H., Han, J., Wu, P., Wang, J., Tu, J., Tang, H., Zhu, L.: Machine learning and human-machine trust in healthcare: A systematic survey. CAAI Transactions on Intelligence Technology9(2), 286–302 (2024)

    2024

  8. [16]

    arXiv preprint arXiv:2402.14035 (2024)

    Liu,Z.,Liu,Q.,Li,Y.,Liu,L.,Shrivastava,A.,Bi,S.,Hong,L.,Chi,E.H.,Zhao,Z.: Wisdom of committee: Distilling from foundation model to specialized application model. arXiv preprint arXiv:2402.14035 (2024)

    2024 arXiv

  9. [17]

    Acta Informatica Medica30(3), 230 (2022)

    Masic, I.: Medical decision making-an overview. Acta Informatica Medica30(3), 230 (2022)

    2022

  10. [18]

    Inter- national journal of preventive medicine5(9), 1073 (2014)

    Masic, I., Hodzic, A., Mulic, S.: Ethics in medical research and publication. Inter- national journal of preventive medicine5(9), 1073 (2014)

    2014

  11. [19]

    In: ICML 2026 Workshop on Structured Data for Health

    Mohan, M., Lingisetty, S., Ahmad, U., Kumar, A., Harikrishnan, S., Chamarty, S., Zhu, K.: A fairness audit of medical imaging foundation models on a multimodal structured clinical benchmark. In: ICML 2026 Workshop on Structured Data for Health

    2026

  12. [20]

    arXiv preprint arXiv:2501.15579 (2025)

    Nie, Y., He, S., Bie, Y., Wang, Y., Chen, Z., Yang, S., Cai, Z., Wang, H., Wang, X., Luo, L., et al.: An explainable biomedical foundation model via large-scale concept-enhanced vision-language pre-training. arXiv preprint arXiv:2501.15579 (2025)

    2025 arXiv

  13. [21]

    Patel, D.: Data sovereignty and algorithmic dependency in the global south: A systematic review of foundation model governance, digital inequity, and decolonial artificial intelligence frameworks (2026)

    2026

  14. [22]

    Energies 17(23), 5965 (2024)

    Pimenow, S., Pimenowa, O., Prus, P.: Challenges of artificial intelligence develop- ment in the context of energy consumption and impact on climate change. Energies 17(23), 5965 (2024)

    2024

  15. [23]

    Intelligence-based medicine3, 100005 (2020) 12 J.S.H

    Rezaei, M., Shahidi, M.: Zero-shot learning and its applications from autonomous vehicles to covid-19 diagnosis: A review. Intelligence-based medicine3, 100005 (2020) 12 J.S.H. Baxter & E. Germani

    2020

  16. [24]

    In: Cur- rent controversies in values and science, pp

    Rooney, P.: The borderlands between epistemic and non-epistemic values. In: Cur- rent controversies in values and science, pp. 31–45. Routledge (2017)

    2017

  17. [25]

    arXiv preprint arXiv:2312.10065 (2023)

    Saravanan, A.P., Kocielnik, R., Jiang, R., Han, P., Anandkumar, A.: Exploring social bias in downstream applications of text-to-image foundation models. arXiv preprint arXiv:2312.10065 (2023)

    2023 arXiv

  18. [26]

    In- ternational journal of computer vision128(2), 336–359 (2020)

    Selvaraju, R.R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., Batra, D.: Grad- cam: visual explanations from deep networks via gradient-based localization. In- ternational journal of computer vision128(2), 336–359 (2020)

    2020

  19. [27]

    In: MICCAI Workshop on Fairness of AI in Medical Imaging

    Stanley, E.A., Souza, R., Winder, A.J., Wilms, M., Pike, G.B., Dagasso, G., Nielsen, C., MacEachern, S.J., Forkert, N.D.: Assessing the impact of sociotechni- cal harms in ai-based medical image analysis. In: MICCAI Workshop on Fairness of AI in Medical Imaging. pp. 163–175. S...

    2024

  20. [28]

    Nature Biomedical Engineering9(4), 521–538 (2025)

    Sun, Y., Wang, L., Li, G., Lin, W., Wang, L.: A foundation model for enhanc- ing magnetic resonance images and downstream segmentation, registration and diagnostic tasks. Nature Biomedical Engineering9(4), 521–538 (2025)

    2025

  21. [29]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Tian, L., Greer, H., Kwitt, R., Vialard, F.X., San José Estépar, R., Bouix, S., Rushmore, R., Niethammer, M.: unigradicon: A foundation model for medical im- age registration. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 749–7...

    2024

  22. [30]

    Advances in Neural Information Processing Systems36, 74952–74965 (2023)

    Turpin, M., Michael, J., Perez, E., Bowman, S.: Language models don’t always say what they think: Unfaithful explanations in chain-of-thought prompting. Advances in Neural Information Processing Systems36, 74952–74965 (2023)

    2023

  23. [31]

    arXiv preprint arXiv:2507.06952 (2025)

    Vafa, K., Chang, P.G., Rambachan, A., Mullainathan, S.: What has a founda- tion model found? using inductive bias to probe for world models. arXiv preprint arXiv:2507.06952 (2025)

    2025

  24. [32]

    arXiv preprint arXiv:2207.02852 (2022)

    Villalobos, P., Sevilla, J., Besiroglu, T., Heim, L., Ho, A., Hobbhahn, M.: Ma- chine learning model sizes and the parameter gap. arXiv preprint arXiv:2207.02852 (2022)

    2022 arXiv

  25. [33]

    Scientific Data10(1), 574 (2023)

    Wang, D., Wang, X., Wang, L., Li, M., Da, Q., Liu, X., Gao, X., Shen, J., He, J., Shen, T., et al.: A real-world dataset and benchmark for foundation model adaptation in medical image classification. Scientific Data10(1), 574 (2023)

    2023

  26. [34]

    In: International Conference on Electronic Government and the Information Systems Perspective

    Wójcik, M.A.: Foundation models in healthcare: opportunities, biases and regula- tory prospects in europe. In: International Conference on Electronic Government and the Information Systems Perspective. pp. 32–46. Springer (2022)

    2022

  27. [35]

    Quantitative Science Studies6, 623–651 (2025)

    Xiang, S., Romero, D., Teplitskiy, M.: Research topic choice: Motivations, strate- gies, and consequences. Quantitative Science Studies6, 623–651 (2025)

    2025

  28. [36]

    In: Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society

    Zhang, B.H., Lemoine, B., Mitchell, M.: Mitigating unwanted biases with adver- sarial learning. In: Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society. pp. 335–340 (2018)

    2018

  29. [37]

    ACM Computing Surveys58(11), 1–35 (2026)

    Zhang, Y., Gao, J., Tan, Z., Zhou, L., Ding, K., Zhou, M., Zhang, S., Wang, D.: Data-centric foundation models in computational healthcare: A survey. ACM Computing Surveys58(11), 1–35 (2026)

    2026

  30. [38]

    arXiv preprint arXiv:2502.04386 (2025)

    Zheng, G., Jacobs, M.A., Braverman, V., Parekh, V.S.: Towards fair medi- cal ai: Adversarial debiasing of 3d ct foundation embeddings. arXiv preprint arXiv:2502.04386 (2025)

    2025 arXiv

  31. [39]

    arXiv preprint arXiv:2408.00874 (2024)

    Zhu, J., Hamdi, A., Qi, Y., Jin, Y., Wu, J.: Medical sam 2: Segment medical images as video via segment anything model 2. arXiv preprint arXiv:2408.00874 (2024)

    2024 arXiv

  32. [40]

    arXiv preprint arXiv:2306.15546 (2023)

    Zhuang, W., Chen, C., Li, J., Chen, C., Jin, Y., Lyu, L.: When foundation mod- els meet federated learning: Motivations, challenges, and future directions. arXiv preprint arXiv:2306.15546 (2023)

    2023 arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.