From Manuals to Maintenance: Fine-Tuning MedGemma for Multi-Modal Imaging System Support in Low-Resource Settings

Abdulrazaq Zubair; Ahmed Tahiru Issah; Alyasaa Anas; Bernes Lorier Atabonfack; Clemence Ingabire; Maruf Adewole; Mawuli Deku; Mohammed Hardi Abdul Baaki; Raymond Confidence; Tolulope Olusuyi

REVIEW 4 major objections 5 minor 14 references

From Manuals to Maintenance: Fine-Tuning MedGemma for Multi-Modal Imaging System Support in Low-Resource Settings

T0 review · 4 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Fine-tuning MedGemma on manual-derived QA pairs substantially improves imaging-equipment troubleshooting accuracy.

desk verdict A genuinely useful new dataset and a sensible fine-tuning study, but the 'procedurally accurate' claim is not yet supported by the AI-generated reference evaluation. read the letter →

arxiv 2608.08896 v1 pith:DT6K5CLH submitted 2026-08-09 cs.AI

Bernes Lorier Atabonfack , Zion Kongbi Nfo , Ahmed Tahiru Issah , Tolulope Olusuyi , Clemence Ingabire , Mohammed Hardi Abdul Baaki , Mawuli Deku , Abdulrazaq Zubair

show 4 more authors

Alyasaa Anas Raymond Confidence Maruf Adewole Udunna C. Anazodo

This is my paper · ORCID

classification cs.AI

keywords MedicalfoundationmodelsMRImaintenanceUltrasoundParameter-efficientfine-tuningQLoRALow-resourcesettingsSyntheticQAdatasetEquipmenttroubleshooting

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 a small, openly available medical language model can be adapted to help biomedical technicians fix MRI and ultrasound machines in low-resource settings. The authors built a dataset, INGENZI_DatasetV1, of 10,294 question-answer pairs generated automatically from manufacturer maintenance manuals, and fine-tuned the 4-billion-parameter MedGemma model on them using quantized low-rank adaptation. Comparing the fine-tuned model to the unmodified base model, they report that token-level F1 rises from 0.22 to 0.38, ROUGE-2 from 0.18 to 0.41, and BERTScore F1 from 0.86 to 0.91. If those numbers hold, a lightweight model could deliver step-by-step repair instructions without internet or specialized OEM support, addressing a documented cause of imaging equipment downtime across nine LMICs.

What carries the argument

The load-bearing objects are two: the INGENZI_DatasetV1 corpus, a set of 10,294 QA-context triples produced by chunking 2,077 pages of Siemens MRI manuals and 2,322 pages of Philips ultrasound manuals into 512-character windows with a 64-character overlap and running an automated generator at temperature 0.0 followed by rule-based and manual filtering; and QLoRA fine-tuning, which freezes a 4-bit-quantized MedGemma-4b-it and trains low-rank adapters of rank 16 on its linear layers. Their joint role is to teach the model the local vocabulary of error codes, calibration steps, and part replacements without requiring full fine-tuning hardware.

What would settle it

Have two or more biomedical engineers independently score the fine-tuned model's answers on real error-log queries drawn from manuals never used in training, with the base model and the engineers' own repair orders as comparisons; if engineer-rated correctness is indistinguishable from baseline, or if overlapping 512-character chunks from the same manual appear in both training and test sets, the claimed procedural gains do not hold.

Watch

Extended reading notes

Core claim

The paper claims that domain-specific fine-tuning of MedGemma-4b-it on INGENZI_DatasetV1 converts a general medical assistant into a procedurally accurate maintenance advisor for Siemens MAGNETOM MRI and Philips HDI 5000 ultrasound systems. The reported metric gains over the base model are F1 0.22 to 0.38, ROUGE-1 0.37 to 0.57, ROUGE-2 0.18 to 0.41, ROUGE-L 0.26 to 0.48, and BERTScore F1 0.86 to 0.91. The authors interpret these gains as evidence that the model now reproduces structured, stepwise technical instructions rather than generic references to a manual. The system is a causal language model prompted with instruction-style question-answer examples, trained for 15 epochs with QLoRA adapters, and evaluated without retrieval augmentation.

Load-bearing premise

The central claim rests on the assumption that the 10,294 AI-generated question-answer pairs, filtered only by automated rules and quick manual review, are correct enough to serve as ground truth, and that the random train/test split does not let overlapping manual chunks leak answers into evaluation.

Editorial extensions

If this is right

The fine-tuned MedGemma model can be run locally as a decision-support tool for technicians, since QLoRA preserves the base model's small footprint.
A future hybrid system can add retrieval augmentation on top of this fine-tuned core to enforce factual grounding, a route the authors explicitly flag.
The open INGENZI_DatasetV1 corpus makes the pipeline reproducible and extendable to other modalities and OEM documentation.
The reported metric pattern implies that even before retrieval is added, the model captures technical phrasing well enough to serve as a strong baseline for maintenance QA.

Reading between the lines

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

If the synthetic QA labels are approximately correct, the same chunk-and-generate pipeline could be applied to ventilators, X-ray systems, or laboratory analyzers, provided a domain expert reviews a sample of questions and answers first.
The gap between BERTScore (0.86 to 0.91) and lexical F1 (0.22 to 0.38) suggests the model retains semantic gist but rarely reproduces exact wording; a deployed assistant would need human confirmation before any repair step is attempted.
Because the data are split randomly with overlapping 512-character chunks, some test questions may share nearly identical context with training questions; a manual-level or device-level split would give a stiffer test of generalization.

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.

Desk Editor's Note

read the letter

Worth a look: the INGENZI_DatasetV1 is a genuinely new public resource—about 10k QA pairs derived from Siemens MRI and Philips ultrasound service manuals—and the paper gives a clear account of a standard QLoRA fine-tuning pipeline on MedGemma-4B. The survey of 61 facilities across nine LMICs is a nice addition; the numbers on documentation gaps and downtime are useful context. If you work on medical ML in low-resource settings, the dataset alone is a reason to engage.

The soft spots are all in the evaluation. The reference answers were generated by LlamaIndex's RagDatasetGenerator from 512-character chunks, with heuristic filtering but no expert review. Every reported gain is measured against those AI-written references. So a ROUGE-2 jump from 0.18 to 0.41 may largely reflect the model learning to imitate the generator's phrasing, not learning correct maintenance procedures. A plausible-sounding but wrong step can score well on lexical and semantic overlap. The abstract's "procedurally accurate" overstates what the metrics can show. To the authors' credit, they list expert validation as a prerequisite in Section 6—so they know.

Two smaller issues. The chunks overlap by 64 characters, and the train/test split is random, so the same passage can appear on both sides; this likely inflates all numbers to some degree. And the BERTScore figure is reported as 0.91 in the abstract and Table 2 but 0.92 in the Section 4 text; that kind of inconsistency makes a careful reader wonder about the other numbers. No error bars or statistical tests, either. No code is released, which makes the pipeline harder to reproduce.

None of this kills the dataset or the basic finding that fine-tuning helps. But the central claim needs better support: a non-overlapping split, expert evaluation of a sample of outputs by biomedical engineers, and a corrected, consistent set of numbers. As it stands, this is a useful dataset-and-baseline paper, not a demonstration of deployment-ready accuracy.

I would send it to peer review—the dataset fills a real gap—but I'd ask for major revision on those evaluation points before accepting.

Referee Report

4 major / 5 minor

Summary. The paper presents INGENZI_DatasetV1, a synthetic question-answer dataset built from Siemens MRI and Philips ultrasound manuals, and uses QLoRA to fine-tune MedGemma-4b-it for equipment troubleshooting QA. The authors report substantial metric improvements over the base model (F1 0.22 to 0.38, ROUGE-2 0.18 to 0.41, BERTScore F1 0.86 to 0.91) and interpret these as evidence of more precise and procedurally accurate technical responses. The manuscript also reports a nine-country survey of imaging-equipment maintenance practices and publicly releases the dataset. The evaluation, however, is entirely automated and the reference answers are themselves LLM-generated from the same manual corpus, which is acknowledged in Section 6.

Significance. If the headline gains were validated against expert-derived ground truth, this work would be a useful step toward low-cost, deployable technical support for imaging equipment in resource-limited settings. The public dataset and the reproducible QLoRA fine-tuning pipeline are concrete contributions, and the multi-country survey is valuable context for the problem. The central weakness is that all quantitative claims rest on synthetic references with no expert adjudication, so the paper's stronger conclusions about procedural accuracy outrun the evidence. Confirming or reframing the claims with a modest expert evaluation would materially change the significance of the work.

major comments (4)

[Abstract; Section 4; Section 6] The claim that fine-tuning yields 'significantly more precise and procedurally accurate technical responses' is not supported by the evaluation, because every metric in Table 2 compares model outputs against reference answers generated by LlamaIndex RagDatasetGenerator from the same manual chunks, filtered only by heuristics. Section 6 explicitly states that performance was 'assessed solely via automated metrics' and that expert validation 'remains a prerequisite for real-world deployment.' Lexical and embedding overlap with chunk-derived references is weak evidence of procedural correctness: a plausible but wrong repair step can score well on ROUGE and BERTScore. At minimum, the authors should sample a few hundred test examples for review by biomedical engineers and report agreement, or substantially soften the wording from 'procedurally accurate' to 'lexically closer to generated references.'
[Section 3.2.2; Section 3.3] The dataset construction uses 512-character chunks with a 64-character overlap, and the train/validation/test split is a random split on samples. Because overlapping chunks from the same manual page can appear in both training and test partitions, the reported improvements may be inflated by near-duplicate text at train/test boundaries. The authors should split at the document or page level, or deduplicate overlapping chunks, and re-run the evaluation to rule out leakage.
[Section 4; Table 2] The reported BERTScore F1 is inconsistent: the abstract and Table 2 report 0.86 to 0.91, while the Section 4 text states 'BERTScore F1 increased from 0.86 to 0.92.' In addition, Table 2 gives single point estimates with no confidence intervals, standard deviations, or significance tests, so the word 'significantly' in the abstract is not backed by any statistical evidence. The authors should correct the inconsistency and provide error bars or per-sample distributions for at least the main metrics.
[Section 3.2.2; Section 5] The manual and rule-based filtering step is not specified in enough detail to be reproduced. Terms such as 'insufficient context grounding', 'likely hallucinated answers', and 'generic responses' need operational definitions, and the authors should state whether the manual filtering was performed by one person or multiple annotators and with what instructions. This matters because the filtering directly determines the quality of the synthetic ground truth on which the central evaluation rests.

minor comments (5)

[References] References [9] and [11] are the same QLoRA paper and should be merged or cross-referenced to avoid redundancy.
[Section 3.2.2] The dataset link is given only as 'here [4]' in the text; the actual Hugging Face URL appears only in the reference list. Please make the hyperlink explicit at the point of mention.
[Section 3.3] The learning rate is written as '2×10 −4' with an awkward spacing; please typeset it as 2 × 10^−4 consistently.
[Figures] Figures 1–3 are referenced but not present in the manuscript text provided; please ensure the figures are included in the camera-ready version and that their captions fully describe the workflow and survey results.
[Abstract] The phrase 'multi-modality medical equipment maintenance question-answering framework' is used, but the dataset and experiments cover only MRI and ultrasound; consider clarifying that 'multi-modality' refers to imaging modalities, not input modalities.

Circularity Check

1 steps flagged · score 6.0 of 10

Reported 'procedural accuracy' gains reduce to matching an AI-generated reference distribution that also supplied the training targets.

fitted input called prediction [Abstract; Section 3.2.2; Table 2; Section 6]
"Using RagDatasetGenerator, we generated question, answer, and reference context triples.... These metric gains demonstrate that the model generates significantly more precise and procedurally accurate technical responses to new troubleshooting queries. ... model performance was assessed solely via automated metrics, and rigorous expert human validation by biomedical engineers remains a prerequisite for real-world deployment."

The fine-tuning targets are the RagDatasetGenerator answers, and the evaluation references in Table 2 are drawn from that same AI-generated distribution. ROUGE-2/F1/BERTScore gains therefore measure how well the fine-tuned model imitates the answer generator, not whether the generated steps are procedurally correct. Calling the metric gains a demonstration of 'procedurally accurate technical responses' equates procedural accuracy with agreement with the synthetic references used as training targets. The paper's own Section 6 concedes that only automated metrics were used and expert validation is still required, so the headline claim outruns the evidence and is circular in its operational definition of accuracy.

full rationale

The core derivation is: manuals -> RagDatasetGenerator synthetic QA -> QLoRA fine-tuning on those QA pairs -> metric gains on a held-out split of the same synthetic QA distribution -> 'procedurally accurate' responses. The held-out test answers are not literally in the training set, so this is not a case where the test score is mathematically forced. However, the inference from metric gains to procedural accuracy is circular because 'procedural accuracy' is operationalized solely as agreement with the same AI-generated references that define the training objective. The paper is transparent about this in Section 6 ('assessed solely via automated metrics... expert human validation... prerequisite'), which lowers the severity, but the abstract's 'demonstrate' claim still presents self-agreement as external validity. Additionally, the 64-character overlap between chunks and the random split create leakage risk that can inflate the reported gains. No other circularity is present: the self-citations to the dataset and prior INGENZI Tech work are not load-bearing for the metric computation, and QLoRA and the evaluation metrics are standard external tools. The central weakness is therefore a self-referential evaluation rather than a fully circular derivation, giving a partial circularity score of 6.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim depends on several unverified premises: the correctness and legality of the manuals, the validity of AI-generated QA labels as ground truth, and the interpretability of lexical metrics as evidence of procedural accuracy. No free parameters are fit to the test set, but pipeline hyperparameters shape the dataset and training.

free parameters (4)

chunk_size = 512 characters (64 overlap)
Chosen for the text segmentation step; overlapping chunks may leak context between train and test splits.
generation_temperature = 0.0
Set to reduce hallucination during QA synthesis; influences the style and grounding of all labels.
LoRA_rank = 16
Chosen for QLoRA efficiency; affects adaptation capacity, though not fitted to the test metric.
training_epochs = 15
Chosen based on evaluation every 50 steps; the final checkpoint selection is not described, so the reported metrics may correspond to a particular epoch.

assumptions (3)

domain assumption Source manuals are accurate, current, and legally usable for derivative data creation.
Section 3.2.1 states manuals came from open-access repositories and were used strictly for dataset construction; their correctness is assumed.
ad hoc to paper Filtered AI-generated QA pairs constitute correct ground truth.
Section 3.2.2 assumes filtering removes all hallucinations and errors, with no independent expert audit.
domain assumption ROUGE, F1, and BERTScore are meaningful proxies for repair instruction quality.
Section 4 uses these metrics as sole evidence of procedural accuracy.

how reviews work

0 comments

Cite this review

Pith. "Pith review of From Manuals to Maintenance: Fine-Tuning MedGemma for Multi-Modal Imaging System Support in Low-Resource Settings." pith.science (2026). https://pith.science/paper/DT6K5CLH

@misc{pith2026260808896,
  author       = {Pith},
  title        = {Pith review of: From Manuals to Maintenance: Fine-Tuning MedGemma for Multi-Modal Imaging System Support in Low-Resource Settings},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DT6K5CLH}},
  note         = {Machine review of arXiv:2608.08896}
}

read the original abstract

Imaging device downtime is a major barrier to healthcare delivery in low- and middle-income countries (LMICs), often driven by limited access to specialized biomedical engineering support. We present a multi-modality medical equipment maintenance question-answering (QA) framework and demonstrate the fine-tuning of a medical foundation model for specialized technical troubleshooting tasks. Guided by a multi-country survey across nine LMICs, we curated technical manuals from MRI and ultrasound systems to generate the INGENZI_DatasetV1, containing 10,294 high-quality, filtered QA-context pairs. Using QLoRA-based parameter-efficient fine-tuning, we adapted the MedGemma-4b-it model to interpret system error logs and generate step-by-step equipment repair instructions. Compared to the baseline model, the fine-tuned system achieved substantial improvements across metrics, including F1 score (0.22 to 0.38), ROUGE-2 (0.18 to 0.41), and BERTScore F1 (0.86 to 0.91). These metric gains demonstrate that the model generates significantly more precise and procedurally accurate technical responses to new troubleshooting queries. This work establishes a reliable foundation for AI-assisted diagnostic and maintenance tools in resource-constrained settings.

Figures

Figures reproduced from arXiv: 2608.08896 by the authors.

**Figure 1.** Workflow for constructing the medical imaging equipment QA dataset from technical manuals. 3.3 Model Fine-Tuning and Experimental Setup Using the INGENZI dataset, we fine-tuned the MedGemma-4b-it model for technical question answering using parameter-efficient fine-tuning. The task formulation was causal language modeling, where each training example was formatted into an instruction-style prompt containing the qu… view at source ↗

**Figure 2.** Survey Responses by healthcare facility type (n = 61). In terms of documentation, access to technical documentation was limited, with 41% facilities relying exclusively on hard-copy manuals, 11% on digital manuals, while 33% reported no manuals available, and 15% had both hard-copy and digital formats. Despite these limitations, 72% (44/61) expressed willingness to share technical documentation for knowledge-buildin… view at source ↗

**Figure 3.** Summary of survey findings across 61 healthcare facilities in nine LMICs. A total of 63 non-functional imaging devices were reported, corresponding to approximately one broken device per affected site. Facilities with structured preventive maintenance exhibited substantially lower mean downtime (14.1 days) compared to reactive maintenance settings (40.7 days), representing a 188.9% increase in downtime without str… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

14 extracted references · 7 canonical work pages

[1]

Medical Equipment Maintenance Market,

“Medical Equipment Maintenance Market,” Market.us, Jan. 10, 2025. https://market.us/report/medical-equipment-maintenance-market/ (accessed Feb. 26, 2026)

work page 2025
[2]

Global Atlas of medical devices,

WHO, “Global Atlas of medical devices,” www.who.int, Aug. 05, 2017. https://www.who.int/publications/i/item/9789241512312

arXiv 2017
[3]

Medical equipment donation in low-resource settings: a review of the literature and guidelines for surgery and anaesthesia in low-income and middle-income countries

Marks IH, Thomas H, Bakhet M, Fitzgerald E. Medical equipment donation in low-resource settings: a review of the literature and guidelines for surgery and anaesthesia in low-income and middle-income countries. BMJ Glob Health. 2019 Sep 29;4(5):e001785. doi: 10.1136/bmjgh-2019-001785. PMID: 31637029; PMCID: PMC6768372

work page doi:10.1136/bmjgh-2019-001785 2019
[4]

Hugging Face (2026)

Atabonfack, B.L., Issah, A.T., Baaki, M.H.A., Ingabire, C., Olusuyi, T., Kongbi, Z., Adewole, M., Anazodo, U.C.: Imaging Equipment Technical QA Dataset. Hugging Face (2026). https://huggingface.co/datasets/BernesLy/INGENZI_DatasetV1

work page 2026
[5]

L., Issah, A

Atabonfack, B. L., Issah, A. T., Baaki, M. H., Ingabire, C., Olusuyi, T., Ade- wole, M., Anazodo, U. C., & Brown, T. X. (2026). Empowering Medical Equip- ment Sustainability in Low-Resource Settings: An AI-Powered Diagnostic and Support Platform for Biomedical Technicians. Nature, 217-230, 978-3-032-13654. "https://link.springer.com/chapter/10.1007/978-3-...

work page doi:10.1007/978-3-032-13654-1_22 2026
[6]

MedGemma Technical Report

Sellergren et al. "MedGemma Technical Report." arXiv preprint arXiv:2507.05201 (2025)

arXiv 2025
[7]

Singhal et al.,Med-PaLM, Nature, 2023

work page 2023
[8]

LLaVA-Med Li et al., NeurIPS Workshop, 2023

work page 2023

Show all 14 references

[10]

S., Wei, J., Chung, H

Singhal, K., Azizi, S., Tu, T., Mahdavi, S. S., Wei, J., Chung, H. W., Scales, N., Tanwani, A., Cole-Lewis, H., Pfohl, S., Payne, P., Seneviratne, M., Gamble, P., Kelly, C., Babiker, A., Schärli, N., Chowdhery, A., Mansfield, P., Demner-Fushman, D., . . . Natarajan, V. (2023)....

2023 doi
[11]

Dettmers, T., Pagnoni, A., Holtzman, A., & Zettlemoyer, L. (2023). QLoRA: Ef- ficient Finetuning of Quantized LLMs. ArXiv. https://arxiv.org/abs/2305.14314

2023 arXiv
[12]

C., Bhaduri, S., Roy, T., Jain, V., & Chadha, A

Balne, C. C., Bhaduri, S., Roy, T., Jain, V., & Chadha, A. (2024). Parame- ter Efficient Fine Tuning: A Comprehensive Analysis Across Applications. ArXiv. https://arxiv.org/abs/2404.13506

2024 arXiv
[13]

J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., & Chen, W

Hu, E. J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., & Chen, W. (2021). LoRA: Low-Rank Adaptation of Large Language Models. ArXiv. https://arxiv.org/abs/2106.09685

2021 arXiv
[14]

Q., & Artzi, Y

Zhang, T., Kishore, V., Wu, F., Weinberger, K. Q., & Artzi, Y. (2019). BERTScore: Evaluating Text Generation with BERT. arXiv preprint arXiv:1904.09675. https://doi.org/10.48550/arxiv.1904.09675

2019 doi arXiv
[15]

Lin, C. Y. (2004). ROUGE: A Package for Automatic Evaluation of Summaries. In Text Summarization Branches Out (pp. 74–81). Barcelona, Spain: Association for Computational Linguistics. https://aclanthology.org/W04-1013/

2004

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

Medical Equipment Maintenance Market,

“Medical Equipment Maintenance Market,” Market.us, Jan. 10, 2025. https://market.us/report/medical-equipment-maintenance-market/ (accessed Feb. 26, 2026)

work page 2025

[2] [2]

Global Atlas of medical devices,

WHO, “Global Atlas of medical devices,” www.who.int, Aug. 05, 2017. https://www.who.int/publications/i/item/9789241512312

arXiv 2017

[3] [3]

Medical equipment donation in low-resource settings: a review of the literature and guidelines for surgery and anaesthesia in low-income and middle-income countries

Marks IH, Thomas H, Bakhet M, Fitzgerald E. Medical equipment donation in low-resource settings: a review of the literature and guidelines for surgery and anaesthesia in low-income and middle-income countries. BMJ Glob Health. 2019 Sep 29;4(5):e001785. doi: 10.1136/bmjgh-2019-001785. PMID: 31637029; PMCID: PMC6768372

work page doi:10.1136/bmjgh-2019-001785 2019

[4] [4]

Hugging Face (2026)

Atabonfack, B.L., Issah, A.T., Baaki, M.H.A., Ingabire, C., Olusuyi, T., Kongbi, Z., Adewole, M., Anazodo, U.C.: Imaging Equipment Technical QA Dataset. Hugging Face (2026). https://huggingface.co/datasets/BernesLy/INGENZI_DatasetV1

work page 2026

[5] [5]

L., Issah, A

Atabonfack, B. L., Issah, A. T., Baaki, M. H., Ingabire, C., Olusuyi, T., Ade- wole, M., Anazodo, U. C., & Brown, T. X. (2026). Empowering Medical Equip- ment Sustainability in Low-Resource Settings: An AI-Powered Diagnostic and Support Platform for Biomedical Technicians. Nature, 217-230, 978-3-032-13654. "https://link.springer.com/chapter/10.1007/978-3-...

work page doi:10.1007/978-3-032-13654-1_22 2026

[6] [6]

MedGemma Technical Report

Sellergren et al. "MedGemma Technical Report." arXiv preprint arXiv:2507.05201 (2025)

arXiv 2025

[7] [7]

Singhal et al.,Med-PaLM, Nature, 2023

work page 2023

[8] [8]

LLaVA-Med Li et al., NeurIPS Workshop, 2023

work page 2023