Pith. sign in

REVIEW 3 major objections 5 minor 29 references

Retrieval-Augmented Interpretable Learning: Towards Task-Specific Zero-Shot Models in Healthcare

T0 review · 3 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read A natural-language description plus a memory of prior task predictors is enough to synthesize an interpretable linear clinical model, reaching about 73% accuracy with no task-specific labels.

desk verdict RAIL is a genuinely novel probabilistic framework for zero-shot interpretable predictors, but the headline 73.4% zero-shot accuracy is undermined by a likely patient-level overlap between source-task training and target-task evaluation that the paper never rules out. read the letter →

arxiv 2607.17508 v2 pith:DZWYDSL5 submitted 2026-07-20 cs.LG cs.AI

Sazan Mahbub , Caleb Ellington , Zhiyuan Li , Yixin Yang , Souvik Kundu , Ben Lengerich , Eric P. Xing This is my paper
classification cs.LGcs.AI
keywords zero-shotmodelgenerationretrieval-augmentedlearningmeta-learninginterpretableclinicalpredictioncoefficient-spacetransferuncertaintyquantificationlong-tailedtaskstask-specificlinearmodels
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

RAIL claims that a new clinical procedure can be given an interpretable linear predictor without collecting task-specific labels, as long as there is a natural-language description of the procedure and a memory of prior procedures with text embeddings and trained linear coefficients. The method retrieves the closest prior tasks, mixes their coefficients through an attention mechanism, and produces a distribution over the new task's coefficients; with no labels, the center of that distribution is the deployed model, and with 2-4 labels a revised distribution interpolates between it and a task-specific fit. It reports 73.4% accuracy on held-out zero-shot procedures and roughly 73.2% accuracy with 2-4 examples, while task-specific logistic regression falls to about 55% accuracy in the same few-shot regime. The paper's own ablations show the gain depends on meaningful retrieval: random retrieval drops zero-shot accuracy to 65.9% and semantically distant retrieval to 58.7%. For a sympathetic reader, the core claim is that task text plus a library of prior predictors can synthesize an inspectable, uncertainty-aware model for the long tail of healthcare tasks.

What carries the argument

The load-bearing object is the parametric memory: task embeddings as semantic keys and logistic-regression coefficients as values. Given a target description, a cosine-similarity retrieval prior defines a candidate set; a learned categorical retriever, regularized toward that prior by a KL term, produces retrieval weights; and a multi-head cross-attention generator reads the retrieved coefficient vectors and emits the mean and diagonal variance of a Gaussian coefficient prior. Variational inference with a weighted ELBO ties the pieces together: the likelihood term trains generated coefficients to predict labels, the coefficient KL anchors the posterior to the retrieved prior, and the retriev

What would settle it

Split the clinical database so that no patient appears in both the source-task training admissions and the zero-shot evaluation admissions, then rerun the held-out zero-shot protocol. If accuracy falls noticeably below the reported 73.4%, part of the claimed transfer comes from overlapping patients rather than from coefficient transfer between semantically related tasks; if accuracy stays near 73%, the retrieval mechanism is doing the work.

Watch

Extended reading notes

Core claim

The central claim, on the paper's own terms, is that coefficient-space transfer is a viable route to zero-shot clinical prediction. Each stored task is a pair: a text embedding of the procedure description and the coefficient vector of a logistic-regression model fit on shared diagnostic features. For a new task described only in text, RAIL uses cosine similarity to select candidate source tasks, a learned retriever to re-weight them, and a multi-head attention generator to synthesize a Gaussian prior over the target coefficients. In zero-shot deployment, the prior mean is the model; in few-shot mode, the posterior mean interpolates between this prior and a task-specific logistic-regression

Load-bearing premise

The whole mechanism rests on the assumption that procedures whose text descriptions are close in the embedding space also need similar coefficients in the 217-feature diagnostic space; the paper verifies this only within one hospital system's procedure list, and its own ablations show that random retrieval drops accuracy from 73.4% to 65.9%, so the semantic-retrieval assumption is doing real work.

Editorial extensions

If this is right

  • Newly defined or rare procedures can receive a deployed, inspectable predictor immediately from their text description, without waiting for labels to accumulate.
  • With only 2-4 labeled examples, RAIL keeps roughly 73% accuracy while task-specific logistic regression degrades to about 55%, so the method's value is concentrated exactly in the low-data tail.
  • The method's success is bounded by the quality of the task-text embedding and the coverage of the memory; clinically informed embeddings outperform generic ones, and random retrieval collapses accuracy.
  • RAIL's uncertainty estimates can support selective prediction: keeping only low-uncertainty predictions reduces risk as coverage shrinks, enabling deferral of uncertain cases to clinicians.
  • Coefficient-level posterior intervals distinguish stable from unstable feature explanations, making the interpretability itself uncertainty-aware.

Reading between the lines

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

  • A natural extension the paper leaves implicit: the same recipe should apply to any domain with a fixed feature space and a long-tailed label distribution, such as medication orders, imaging follow-ups, or non-clinical operational tasks, since nothing in the mechanism is procedure-specific.
  • Because the generator outputs a distribution over coefficients rather than a point model, one testable extension is to use multiple sampled models as an ensemble at prediction time; the paper uses sampling mainly for uncertainty, not for accuracy gains.
  • A consequence worth checking outside the reported data: if the text encoder is changed but still clinically informed, retrieval should stay useful as long as the embedding space preserves procedure semantics; the paper's compression experiments hint at this but do not test reordering or retraining the encoder.
  • The selective-prediction result suggests a clinician-in-the-loop deployment protocol, but the paper does not measure the cost of false flags or the clinician time needed to review them.
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 introduces RAIL, a probabilistic meta-learning framework that generates task-specific interpretable linear predictors in a zero-shot manner from a natural-language task description and a memory of previously learned task-specific coefficients. A frozen text encoder embeds task descriptions; a learned retriever selects source tasks by cosine similarity plus a learned score; a multi-head cross-attention generator maps the retrieved coefficient values to a Gaussian prior over the target coefficients. Training uses an ELBO objective with retrieval and coefficient KL regularizers. The method is evaluated on MIMIC-IV clinical procedure prediction tasks across supervised, few-shot, and zero-shot regimes, reporting 73.4% zero-shot accuracy and 73.2% few-shot accuracy, with a range of ablations and uncertainty diagnostics. The authors also provide an ELBO derivation and a Rademacher-complexity generalization bound in the appendix.

Significance. If the empirical claims hold, RAIL is a meaningful contribution: it combines retrieval-augmented generation, meta-learning, and interpretable linear models in a way that produces inspectable, uncertainty-aware predictors for new clinical tasks without task-specific labels. The paper's strengths include a carefully specified probabilistic model, a complete ELBO derivation (Appendix F.1), a correct Rademacher bound (Appendix F.2), extensive ablations (retrieval perturbations, text encoders, embedding corruption), and uncertainty-based selective prediction analyses. The central claim, however, rests on the validity of the zero-shot evaluation and on the absence of patient-level information leakage between source-task training sets and target-task evaluation sets. The manuscript does not currently establish this, and the abstract's characterization of the few-shot supervised baseline as 'close to chance' is not supported by the reported numbers. These issues are load-bearing for the headline result, so the contribution is significant but conditional on a re-evaluation with proper patient-level splitting.

major comments (3)
  1. [§3.1, Appendix B] Patient-level leakage between source-task training and zero-shot evaluation is not ruled out. The dataset construction in Appendix B states that all tasks are built from the same admissions and that each task is split 50/50 within the task, but it does not partition admissions by patient identity. A single admission can have multiple procedures, so an admission used for zero-shot evaluation of target procedure A can appear in the training set of a retrieved source procedure B. The main text says 'target tasks are completely excluded from the retrieval memory: their descriptions, patient samples, labels, and task-specific coefficients are unseen during inference,' but this only excludes the target task's own samples from the memory; it does not prevent the same patients from entering the memory through other tasks' training sets. If such overlap exists, the reported 73.4% zero-shot accura
  2. [Abstract vs. Table 3] The abstract states that in the extreme few-shot regime (2–4 examples) 'supervised task-specific models perform close to chance.' This is contradicted by Table 3, which reports a task-specific LR oracle accuracy of 0.550 in the 2–4 regime, substantially above the 0.5 chance level, and an F1 of 0.424. While F1 is weaker, accuracy is not 'close to chance.' This overstatement appears in the paper's headline comparison and should be corrected; otherwise the claimed advantage of RAIL in the few-shot regime is misrepresented.
  3. [Appendix F.2, Eq. (33)] The generalization bound in Proposition 1 assumes the retrieval-conditioned prior mean µ_phi,t is fixed independently of the target-task sample. If, as described in Appendix B, the source-task training sets that determine µ_phi,t contain the same admissions as the target-task evaluation sample, this independence assumption is violated: the prior is then constructed from features of exactly the units being predicted. The bound therefore does not apply to the actual zero-shot evaluation under the current data construction. This is not an error in the Rademacher argument itself, but it is an omitted condition that must be stated and satisfied for the bound to support the paper's generalization claims.
minor comments (5)
  1. [§2.3–2.4] The retrieval hyperparameters S, k, tau_p, and tau_q appear in Eqs. (4), (8), and (9) but are never given concrete values in the main text or appendix. Since k and S control the retrieval mechanism and its computational cost, reporting their values would improve reproducibility.
  2. [Figure 1 caption] The caption contains garbled fragments such as 'Task emb. th' and 'Features and labels th Task pool'; these appear to be LaTeX artifacts and should be cleaned.
  3. [§3.2] The sentence 'RAIL remains comparatively stable, with F1 between 0.70 and 0.75' is not precisely consistent with Table 3, which reports F1=0.7129 in the 2–4 regime; the range is technically correct but the wording implies a tighter band. Clarify the intended claim.
  4. [Appendix F.1] The ELBO derivation is sound, but the phrase 'the exact set-valued view' followed by 'the closed-form categorical KL below corresponds to this implemented relaxation' is slightly confusing. A sentence clarifying that the objective is optimized over the categorical relaxation, not the set-valued distribution, would help the reader avoid misinterpretation.
  5. [§3.8] The limitations paragraph acknowledges dependence on task-memory coverage and the associational nature of the linear explanations. This is helpful, but it could be more specific: the paper does not characterize how performance degrades as the number of source tasks shrinks, nor does it test retrieval on out-of-domain task descriptions. A sentence noting this would be useful.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: zero-shot targets are excluded from the memory, generated coefficients are a forward-pass function of description and retrieved source coefficients, not a fit to target labels.

full rationale

RAIL's zero-shot inference is not circular. Target tasks in zero-shot evaluation are explicitly excluded from the retrieval memory ('their descriptions, patient samples, labels, and task-specific coefficients are unseen during inference'), and the synthesized model is the prior mean μ_φ,t produced by a fixed generator-retriever from the target description and retrieved source entries. Neither retrieval weights nor generated coefficients are fitted to target labels, so the reported 73.4% accuracy is a genuine held-out evaluation rather than a refit. Source-task coefficients are inputs to a learned synthesis, not renamed target predictions; the comparison with Top-k averaging and retrieval-free baselines shows the synthesis adds content. The paper's self-citations [2,3,4] are motivational framing, not load-bearing uniqueness theorems or ansatz imports. The ablation diagnostics (random/distant retrieval) and the external MIMIC-IV benchmark make the central mechanism independently testable. The strongest concern in the manuscript—patient overlap between source and target admissions—is a data-leakage/correctness risk, not a definitional circularity, and is outside the circularity scoring rubric. No step reduces by construction to its input.

Assumptions & free parameters 4 free parameters · 5 assumptions · 1 invented entities

RAIL's central claim is a property of a trained generator/retriever fit via ELBO on 4,412 MIMIC-IV source tasks, plus hand-chosen hyperparameters whose values are not reported (S, k, tau_p, tau_q, beta_theta, beta_I). The framework invokes standard variational and Rademacher machinery, and two domain assumptions validated only in-corpus: frozen-embedding cosine similarity proxies coefficient transferability, and procedure decisions are approximately linear in the 217 lab features. No physical entities are invented; the retrieval latent is a standard variational construct without independent external evidence.

free parameters (4)
  • Generator/retriever neural network weights = not disclosed
    The zero-shot model IS the output of the trained generator; its weights are fit via the negative ELBO on 4,412 source tasks. The central claim (73.4% zero-shot accuracy) is a property of this fitted function, so the fit is the contribution rather than a derived constant.
  • Posterior gate alpha (Eq. 12) = not disclosed
    Per-coordinate learned gate interpolating between the retrieval-conditioned prior and the task-specific estimator in few-shot adaptation; learned from source-task validation.
  • KL weights beta_theta and beta_I (Eq. 30) = not reported
    Hand-chosen weights controlling coefficient-prior and retrieval-prior anchoring; values not stated in main text or appendix, and they directly determine how much the learned retriever can drift from the cosine prior.
  • Retrieval hyperparameters S, k, tau_p, tau_q (Eqs. 4, 8-9) = not reported
    Candidate-set size, top-k size, and temperatures chosen by hand; they control the sharpness of the retrieval distributions and are not disclosed.
assumptions (5)
  • domain assumption Cosine similarity in the frozen task-embedding space is a valid proxy for coefficient transferability
    Enters at Eq. (4) (retrieval prior) and is load-bearing for the retrieval mechanism; validated only by in-corpus diagnostics (Figs. 5-6) and highly sensitive to embedding quality (random retrieval drops zero-shot accuracy from 0.734 to 0.659).
  • domain assumption Procedure-administration decisions are approximately linear in the 217 lab features
    RAIL generates linear GLM predictors (Eq. 3); if the optimal decision boundary is nonlinear in lab values, coefficient-space transfer is misspecified. The paper notes it assumes a shared diagnostic-feature space but does not test nonlinear bases.
  • standard math Standard variational inference machinery (Jensen's inequality, mean-field factorization)
    Appendix F.1 derives the ELBO using the factorized variational family of Eq. (7), the reparameterization trick, and the categorical/Gaussian closed-form KLs; all standard.
  • standard math Standard Rademacher complexity and Lipschitz contraction inequalities
    Appendix F.2, Proposition 1, uses the standard Rademacher generalization theorem and the contraction inequality; the proof is correct but standard.
  • domain assumption Frozen text encoders (MedEmbed-large-v0.1, DistilBERT) encode clinically meaningful task semantics
    Zero-shot transfer relies on the text encoder's representations; the paper tests sensitivity via noise/compression/randomization (Fig. 10) but cannot verify external validity beyond MIMIC-IV procedure descriptions.
invented entities (1)
  • Retrieval latent I_t (set of S source-task memory entries)
    purpose: Latent variable selecting which memory entries condition the coefficient prior (Eqs. 2-6); intended to provide uncertainty over retrieval.
    A variational modeling construct, not a physical entity; it is never directly validated outside the paper, only through downstream accuracy of models conditioned on it. Low-risk because it is standard variational machinery rather than an unexplained explanatory entity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Retrieval-Augmented Interpretable Learning: Towards Task-Specific Zero-Shot Models in Healthcare." pith.science (2026). https://pith.science/paper/DZWYDSL5

@misc{pith2026260717508,
  author       = {Pith},
  title        = {Pith review of: Retrieval-Augmented Interpretable Learning: Towards Task-Specific Zero-Shot Models in Healthcare},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DZWYDSL5}},
  note         = {Machine review of arXiv:2607.17508}
}
read the original abstract

We introduce Retrieval-Augmented Interpretable Learning (RAIL), a probabilistic meta-learning framework for zero-shot generation of task-specific interpretable models that synthesizes coefficient-space structure from natural-language task descriptions and a memory of previously learned task-specific predictors. RAIL retrieves related source tasks, transfers structure through coefficient space, and generates a new predictor in the original diagnostic-feature space, enabling zero-shot and few-shot clinical procedure prediction with feature-level explanations. Its probabilistic formulation provides uncertainty over retrieval, model coefficients, and predictions, supporting reliability-aware deployment: uncertain predictions or unstable explanations can be flagged for additional clinical review rather than treated as automatic decisions. This makes RAIL particularly suited for healthcare settings, where prediction tasks are highly long-tailed, new clinical targets arise frequently, and models must remain inspectable, uncertainty-aware, and compatible with human oversight. Across long-tailed clinical procedure prediction tasks, RAIL maintains reliable performance across data-availability regimes: it achieves 73.4% accuracy in the held-out zero-shot settings, where no supervised task-specific model can be trained, and remains near 73.2% accuracy in the extreme few-shot regime with only 2-4 examples, where supervised task-specific models perform close to chance. RAIL further benefits from clinically informed task representations and yields retrieval, uncertainty, and coefficient-level diagnostics that make model behavior more transparent. These results suggest a path toward scalable clinical prediction systems that can adapt to new tasks while preserving interpretability and reliability.

Figures

Figures reproduced from arXiv: 2607.17508 by the authors.

Figure 1
Figure 1. Overview of RAIL. Retrieved task embeddings and learned linear models from the [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. PGM for RAIL. Observed variables are shaded. The embeddings provide semantic alignment across tasks, while the coefficients provide transferable parameter-level structure in the original feature space. RAIL is primarily designed for zero-shot model synthesis: at inference time, no labeled examples from task t are required. When lim￾ited target-task labels are available, the same formulation supports few-shot posteri… view at source ↗
Figure 3
Figure 3. Long-tailed task distribution. Pro￾cedure tasks are highly imbalanced in sample availability. This long-tail structure makes direct task-specific training unreliable in the low-data tail and motivates retrieval-augmented model generation, where related prior tasks provide parameter-level structure for new or data-scarce tasks [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: RAIL remains stable across sample-scarcity [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 10
Figure 10. Figure 10: These analyses show that Gaussian noise degrades few-shot and zero-shot performance, [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]
Figure 5
Figure 5. Figure 5: Embedding–model alignment. Top-k retrieved tasks are closer to the query and more coherent among themselves than random neighbors. 7 [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Oracle retrieval consistency. Em￾bedding retrieval better matches oracle-relevant tasks than random retrieval and TF-IDF. Embedding–model alignment. We first ask whether nearest neighbors in task-embedding space are also close in coefficient space [PITH_FULL_IMAGE:fig…
Figure 7
Figure 7. Figure 7: Predictive uncertainty diagnostics. (a) Predictive confidence tracks empirical accuracy. (b) Higher predictive uncertainty corresponds to higher error rates [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 6
Figure 6. Figure 6: Oracle retrieval consistency. Em￾bedding retrieval better matches oracle-relevant tasks than random retrieval and TF-IDF. Embedding–model alignment. We first ask whether nearest neighbors in task-embedding space are also close in coefficient space [PITH_FULL_IMAGE:fig…
Figure 5
Figure 5. Figure 5: Embedding–model alignment. Top-k retrieved tasks are closer to the query and more coherent among themselves than random neighbors. (a) Confidence vs. accuracy. (b) Error vs. uncertainty [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 8
Figure 8. Figure 8: Selective prediction. Lower￾uncertainty predictions yield lower risk at re￾duced coverage. We focus on predictive uncertainty for reliability analysis: we sample K = 100 coefficient vectors from the posterior, compute the induced predictive probabilities, and use the b…
Figure 9
Figure 9. Figure 9: Feature-level interpretability with uncertainty. RAIL ranks diagnostic features by posterior [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Additional representation ablations. Noise degrades RAIL performance, compact em [PITH_FULL_IMAGE:figures/full_fig_p014_10.png]
Figure 11
Figure 11. Figure 11: Failure detection using uncertainty. Predictive uncertainty is the strongest signal for [PITH_FULL_IMAGE:figures/full_fig_p014_11.png]
Figure 11
Figure 11. Figure 11: Failure detection using uncertainty. Predictive uncertainty is the strongest signal for [PITH_FULL_IMAGE:figures/full_fig_p015_11.png]
Figure 12
Figure 12. Figure 12: Stable and unstable coefficient examples. Stable tasks exhibit top coefficients with [PITH_FULL_IMAGE:figures/full_fig_p015_12.png]

Review history (2 revisions) →

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

29 extracted references · 3 linked inside Pith

  1. [1]

    Meta-learning in neural networks: A survey.IEEE transactions on pattern analysis and machine intelligence, 44(9):5149–5169, 2021

    Timothy Hospedales, Antreas Antoniou, Paul Micaelli, and Amos Storkey. Meta-learning in neural networks: A survey.IEEE transactions on pattern analysis and machine intelligence, 44(9):5149–5169, 2021

    2021

  2. [2]

    Contextualized machine learning.arXiv preprint arXiv:2310.11340, 2023

    Benjamin Lengerich, Caleb N Ellington, Andrea Rubbi, Manolis Kellis, and Eric P Xing. Contextualized machine learning.arXiv preprint arXiv:2310.11340, 2023

    arXiv 2023

  3. [3]

    Learning to estimate sample-specific transcriptional networks for 7,000 tumors.Proceedings of the National Academy of Sciences, 122(21):e2411930122, 2025

    Caleb N Ellington, Benjamin J Lengerich, Thomas BK Watkins, Jiekun Yang, Abhinav K Adduri, Sazan Mahbub, Hanxi Xiao, Manolis Kellis, and Eric P Xing. Learning to estimate sample-specific transcriptional networks for 7,000 tumors.Proceedings of the National Academy of Sciences, 122(21):e2411930122, 2025

    2025

  4. [4]

    Contextualized policy recovery: Modeling and interpreting medical decisions with adaptive imitation learning

    Jannik Deuschel, Caleb Ellington, Yingtao Luo, Ben Lengerich, Pascal Friederich, and Eric P Xing. Contextualized policy recovery: Modeling and interpreting medical decisions with adaptive imitation learning. InForty-first International Conference on Machine Learning, 2024

    2024

  5. [5]

    Large language models in medicine.Nature medicine, 29(8):1930–1940, 2023

    Arun James Thirunavukarasu, Darren Shu Jeng Ting, Kabilan Elangovan, Laura Gutierrez, Ting Fang Tan, and Daniel Shu Wei Ting. Large language models in medicine.Nature medicine, 29(8):1930–1940, 2023

    1930

  6. [6]

    Guiding pretraining in reinforcement learning with large language models, 2023

    Yuqing Du, Olivia Watkins, Zihan Wang, Cédric Colas, Trevor Darrell, Pieter Abbeel, Abhishek Gupta, and Jacob Andreas. Guiding pretraining in reinforcement learning with large language models, 2023

    2023

  7. [7]

    Lagr-seq: Language-guided reinforcement learning with sample-efficient querying, 2023

    Thommen George Karimpanal, Laknath Buddhika Semage, Santu Rana, Hung Le, Truyen Tran, Sunil Gupta, and Svetha Venkatesh. Lagr-seq: Language-guided reinforcement learning with sample-efficient querying, 2023

    2023

  8. [8]

    Adarefiner: Refining decisions of language models with adaptive feedback, 2024

    Wanpeng Zhang and Zongqing Lu. Adarefiner: Refining decisions of language models with adaptive feedback, 2024

    2024

Show all 29 references
  1. [9]

    Zero-shot robustification of zero-shot models, 2024

    Dyah Adila, Changho Shin, Linrong Cai, and Frederic Sala. Zero-shot robustification of zero-shot models, 2024

    2024

  2. [10]

    Causal discovery with language models as imperfect experts, 2023

    Stephanie Long, Alexandre Piché, Valentina Zantedeschi, Tibor Schuster, and Alexandre Drouin. Causal discovery with language models as imperfect experts, 2023

    2023

  3. [11]

    Discovery of the hidden world with large language models, 2024

    Chenxi Liu, Yongqiang Chen, Tongliang Liu, Mingming Gong, James Cheng, Bo Han, and Kun Zhang. Discovery of the hidden world with large language models, 2024

    2024

  4. [12]

    Health-llm: Personalized retrieval-augmented disease prediction system.arXiv preprint arXiv:2402.00746, 2024

    Mingyu Jin, Qinkai Yu, Dong Shu, Chong Zhang, Lizhou Fan, Wenyue Hua, Suiyuan Zhu, Yanda Meng, Zhenting Wang, Mengnan Du, et al. Health-llm: Personalized retrieval-augmented disease prediction system.arXiv preprint arXiv:2402.00746, 2024. 10

    2024 arXiv

  5. [13]

    Attention is all you need.Advances in neural information processing systems, 30, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017

    2017

  6. [14]

    Mimic-iv, a freely accessible electronic health record dataset.Scientific data, 10(1):1, 2023

    Alistair EW Johnson, Lucas Bulgarelli, Lu Shen, Alvin Gayles, Ayad Shammout, Steven Horng, Tom J Pollard, Sicheng Hao, Benjamin Moody, Brian Gow, et al. Mimic-iv, a freely accessible electronic health record dataset.Scientific data, 10(1):1, 2023

    2023

  7. [15]

    What clinicians want: contextualizing explainable machine learning for clinical end use

    Sana Tonekaboni, Shalmali Joshi, Melissa D McCradden, and Anna Goldenberg. What clinicians want: contextualizing explainable machine learning for clinical end use. InMachine learning for healthcare conference, pages 359–380. PMLR, 2019

    2019

  8. [16]

    Learning sample-specific models with low-rank personalized regression.Advances in Neural Information Processing Systems, 32, 2019

    Ben Lengerich, Bryon Aragam, and Eric P Xing. Learning sample-specific models with low-rank personalized regression.Advances in Neural Information Processing Systems, 32, 2019

    2019

  9. [17]

    Meta-learning approaches for causal discovery in dynamic healthcare and robotics environments.Mesopotamian Journal of Artificial Intelligence in Healthcare, 2025:136–153, 2025

    Hayder Abbood MD. Meta-learning approaches for causal discovery in dynamic healthcare and robotics environments.Mesopotamian Journal of Artificial Intelligence in Healthcare, 2025:136–153, 2025

    2025

  10. [18]

    Metapred: Meta- learning for clinical risk prediction with limited patient electronic health records

    Xi Sheryl Zhang, Fengyi Tang, Hiroko H Dodge, Jiayu Zhou, and Fei Wang. Metapred: Meta- learning for clinical risk prediction with limited patient electronic health records. InProceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, pag...

    2019

  11. [19]

    Metacare++: Meta-learning with hierarchical subtyping for cold-start diagnosis prediction in healthcare data

    Yanchao Tan, Carl Yang, Xiangyu Wei, Chaochao Chen, Weiming Liu, Longfei Li, Jun Zhou, and Xiaolin Zheng. Metacare++: Meta-learning with hierarchical subtyping for cold-start diagnosis prediction in healthcare data. InProceedings of the 45th International ACM SIGIR Conference ...

    2022

  12. [20]

    Retrieval-augmented generation for knowledge-intensive nlp tasks.Advances in Neural Information Processing Systems, 33:9459–9474, 2020

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. Retrieval-augmented generation for knowledge-intensive nlp tasks.Advances in Neural Information Processing Systems,...

    2020

  13. [21]

    Retrieval-augmented generation for large language models: A survey.arXiv preprint arXiv:2312.10997, 2023

    Yunfan Gao, Yun Xiong, Xinyu Gao, Kangxiang Jia, Jinliu Pan, Yuxi Bi, Yi Dai, Jiawei Sun, and Haofen Wang. Retrieval-augmented generation for large language models: A survey.arXiv preprint arXiv:2312.10997, 2023

    2023 arXiv

  14. [22]

    Prism: Enhancing protein inverse folding through fine-grained retrieval on structure-sequence multimodal representations.arXiv preprint arXiv:2510.11750, 2025

    Sazan Mahbub, Souvik Kundu, and Eric P Xing. Prism: Enhancing protein inverse folding through fine-grained retrieval on structure-sequence multimodal representations.arXiv preprint arXiv:2510.11750, 2025

    2025

  15. [23]

    Bioadapt- mrc: adversarial learning-based domain adaptation improves biomedical machine reading comprehension task.Bioinformatics, 38(18):4369–4379, 2022

    Maria Mahbub, Sudarshan Srinivasan, Edmon Begoli, and Gregory D Peterson. Bioadapt- mrc: adversarial learning-based domain adaptation improves biomedical machine reading comprehension task.Bioinformatics, 38(18):4369–4379, 2022

    2022

  16. [24]

    Decision transformer: Reinforcement learning via sequence modeling, 2021

    Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Michael Laskin, Pieter Abbeel, Aravind Srinivas, and Igor Mordatch. Decision transformer: Reinforcement learning via sequence modeling, 2021

    2021

  17. [25]

    Causal reasoning and large language models: Opening a new frontier for causality, 2024

    Emre Kıcıman, Robert Ness, Amit Sharma, and Chenhao Tan. Causal reasoning and large language models: Opening a new frontier for causality, 2024

    2024

  18. [26]

    Lmpriors: Pre-trained language models as task-specific priors, 2022

    Kristy Choi, Chris Cundy, Sanjari Srivastava, and Stefano Ermon. Lmpriors: Pre-trained language models as task-specific priors, 2022

    2022

  19. [27]

    Pre-trained language models for interactive decision-making.Advances in Neural Information Processing Systems, 35:31199–31212, 2022

    Shuang Li, Xavier Puig, Chris Paxton, Yilun Du, Clinton Wang, Linxi Fan, Tao Chen, De-An Huang, Ekin Akyürek, Anima Anandkumar, et al. Pre-trained language models for interactive decision-making.Advances in Neural Information Processing Systems, 35:31199–31212, 2022

    2022

  20. [28]

    Efficient reinforcement learning with large language model priors

    Xue Yan, Yan Song, Xidong Feng, Mengyue Yang, Haifeng Zhang, Haitham Bou Ammar, and Jun Wang. Efficient reinforcement learning with large language model priors. In13th International Conference on Learning Representations Iclr 2025, pages 30818–30842. ICLR, 2025. 11

    2025

  21. [29]

    Respiratory Ventilation, 24–96 Consecutive Hours,

    Sebastian Bordt, Ben Lengerich, Harsha Nori, and Rich Caruana. Data science with llms and interpretable models, 2024. A Related Works Clinical predictive modeling often faces a data-scarcity problem: many clinically relevant tasks have limited supervision, yet the resulting mo...

    2024

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.