Pith. sign in

REVIEW 4 major objections 5 minor 59 references

SkillTFM: Gated Skill Evolution for Training-Free Adaptation of Tabular Foundation Models

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read SkillTFM claims tabular foundation models can adapt to distribution shifts and structural boundary failures purely through an external, gated skill state, with no parameter updates, and reports AUC gains of 0.128 to 0.142 with zero…

desk verdict A coherent system paper on gated skill adaptation for TFMs, with solid ablations but incomplete evaluation and missing thresholds. read the letter →

arxiv 2608.06137 v1 pith:WFQA7QBW submitted 2026-08-06 cs.LG

Yi He , Zhengkang Guan , Anpeng Wu , Peng Cui , Fei Wu , Kun Kuang This is my paper · ORCID
classification cs.LG
keywords tabularfoundationmodelstraining-freeadaptationskillbankgatedevolutionboundaryshiftruntimecertificateselectiverepairelectricitypriceforecasting
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

SkillTFM claims that tabular foundation models can be adapted to distribution shift and other boundary failures without any parameter updates, by attaching an external skill state that decides when to repair the base prediction and when to fall back. The system extracts runtime-observable evidence from the task and the base model, activates candidate repair skills, and executes a repair only when a runtime certificate finds the evidence sufficient, the risk acceptable, and no prior failure route blocks the action. New boundary experiences are converted into reusable skills through a promotion gate that requires measurable improvement, low harm rate on guard tasks, and regression and schema checks. On controlled boundary suites the paper reports AUC gains of 0.128 to 0.142 with zero observed harm, and on a nonlinear boundary it reports raising AUC from 0.699 to 0.898 after skill evolution. If these results hold, SkillTFM offers a pluggable way to expand the reliable operating region of a tabular foundation model without touching its weights.

What carries the argument

The central object is the external skill state S = (E, O, R, G, F, B, P), a structured repair policy held outside the base model. E extracts boundary evidence z_T from feature distribution, label-feature relation, prediction bias, trend/periodicity, missingness pattern, and probe sensitivity. O is a bank of verified repair skills, each declaring required evidence and contraindications. R ranks candidate repairs lexicographically by evidence support, historical reliability, risk, memory similarity, and cost. G and F implement risk guards and fallback rules. B and P store rejected routes and principles that constrain future activation, ranking, and certification. The runtime certificate is the gate that executes a repair only when evidence is sufficient, risk is acceptable, and memory does not block the route, otherwise falling back to the base model. The promotion gate admits candidate edits only when intended improvement is at least epsilon, harm rate on guard tasks is at most alpha, and regression and schema checks pass.

What would settle it

Run SkillTFM on a boundary family whose evidence signature is identical to a family where repair helps, but where the repair action actually hurts test performance. If the runtime certificate still certifies the repair because evidence is sufficient, risk is acceptable, and memory does not block the route, the central claim fails; if it falls back or rejects instead, observe the resulting AUC difference to quantify the evidence-transfer gap.

Watch

Extended reading notes

Core claim

The paper's central claim is that TFM boundary failure is not a global property of a task but a local, evidence-conditioned pattern, and that a verifiable external skill state can expand the model's usable boundary by choosing between certified repair and fallback. The skill state S = (E, O, R, G, F, B, P) couples evidence extraction, a verified repair skill bank, ranking rules, risk guards, fallback rules, rejected-route constraints, and principle memory. At runtime the evidence vector z_T = E(T, M) activates only those repair skills whose applicability conditions match and whose contraindications are absent, then a runtime certificate scans ranked candidates and executes the first one that passes evidence, risk, and memory checks; if none passes, the system returns the base prediction. Validation-gated skill evolution then admits a candidate edit only when it improves intended capability on selection tasks, keeps harm on guard tasks below the tolerated margin, and passes regression and schema checks. The paper argues that this two-gate design is what lets SkillTFM improve AUC by 0.128 to 0.142 across held-out, mixed, and generator/severity transfer settings with zero observed harm, and convert an initially unsupported nonlinear boundary from 0.699 to 0.898 AUC.

Load-bearing premise

The load-bearing premise is that the system can tell from the task's observable features and the base model's own predictions when a repair will help and when it will harm, without ever seeing test labels.

Editorial extensions

If this is right

  • SkillTFM can be attached to different tabular foundation models (TabPFN, TabICL, TabDPT, LimiX) with the same external skill state, with held-out AUC improvements around 0.139 to 0.148 per model.
  • The gating policy keeps intervention selective: fallback rates stay between 0.425 and 0.463 on a benchmark with 60 percent boundary-affected samples, and observed harm remains zero.
  • Skill evolution can add a capability that does not exist in the initial skill bank: the nonlinear boundary rises from 0.699 to 0.898 AUC after a gated promotion.
  • In real electricity-price forecasting, SkillTFM reduces MAE from 53.02 to 25.16 across three cities while modifying only 38.2 percent of predictions and covering 83.4 percent of large-error points.
  • Different LLM optimizers can propose candidate edits through a shared interface, but the local promotion gate decides admission, so proposal is separated from deployment.

Reading between the lines

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

  • The same evidence-conditioned certificate idea could be applied to other foundation-model domains, such as images, text, or time series, whenever a frozen predictor's failures leave observable signatures in input or prediction statistics; the paper's evidence channels are tabular-specific but the gating loop is not.
  • Because the zero-harm result is empirical rather than formal, a stronger test would be to run SkillTFM on boundary families whose evidence signatures are deliberately confounded with safe tasks, to see whether the certificate's evidence-to-repair transfer can be fooled.
  • The zero-harm result depends on the guard-task distribution and the tolerated margin delta; changing delta or adding adversarial guard tasks could expose the trade-off between fallback rate and improvement that the current headline numbers average over.
  • The token cost of skill evolution (roughly 19,000 to 35,000 tokens per optimizer alignment, and about 107,000 for the nonlinear skill) suggests a testable extension: measuring whether a smarter proposer reduces alignment tokens while preserving the promotion gate's harm bound.
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

4 major / 5 minor

Summary. SkillTFM proposes a training-free, external 'skill state' that adapts tabular foundation models (TFMs) to boundary shift conditions by extracting runtime evidence, retrieving candidate repair skills, and executing a repair only when a runtime certificate approves it; otherwise it falls back to the base model. A second loop, validation-gated skill evolution, allows candidate edits from traces or LLM proposers to enter the deployed skill state only after utility-improvement, no-harm, regression, and schema checks. The paper reports AUC improvements of 0.128-0.142 with zero observed harm on controlled boundary suites, a nonlinear-boundary improvement from 0.699 to 0.898 after evolution, cross-model and cross-optimizer transfer, and MAE reductions in real-world electricity-price forecasting while modifying only a fraction of predictions.

Significance. If the claims hold, SkillTFM offers a plausible and useful paradigm: adapting frozen TFMs through a pluggable, auditable skill state rather than parameter updates. The paper has real strengths: controlled boundary-family construction, a clear ablation separating evidence-conditioned retrieval from runtime certification (Table 2A), held-out evaluation on a later electricity-price release with explicit target-overlap avoidance (Table 4), and cross-model/cross-optimizer experiments that give evidence of generality. However, the central safety claim is only partially supported because the runtime evidence representation, certificate thresholds, and promotion-gate constants are never specified, and the zero-harm results are confined to tasks generated from the same boundary-family suite used for skill construction and validation. The claimed significance therefore depends on closing these specification gaps and carefully re-scoping the 'zero harm' language.

major comments (4)
  1. [§Methodology, 'Evidence-Conditioned Boundary Expansion', Eq. (5) and Eq. (7)] The paper never defines the feature representation of z_T = E(T,M), the functional forms of r_sup, r_hist, r_risk, r_mem, and cost in Eq. (7), or the thresholds used by the certificate's 'evidence sufficient / risk acceptable / memory permits' checks. Since the runtime certificate is the only deployment-time safeguard against harmful repair, this omission makes the zero-harm claim unverifiable; in particular, the paper does not show that z_T can distinguish a repairable boundary failure from a deployment where the same observable evidence would lead to harm. This is the central load-bearing link between labeled promotion/guard validation (Eqs. 8-10) and unlabeled runtime decisions, and it must be specified before the safety claim can be evaluated.
  2. [§Experiments, Table 1 and §Methodology, Eqs. (3), (9), (10)] The values of epsilon (promotion improvement threshold), alpha (allowed harm rate), and delta (tolerated degradation margin) are never reported for any experiment. Because Eq. (10) requires G_sel >= epsilon and H_sel <= alpha, and Eq. (3) defines the no-harm guarantee in terms of alpha and delta, the reader cannot assess whether the no-harm condition is tight or vacuous; the manuscript should report these values and, if they are chosen by the authors, describe the selection procedure and report sensitivity to them.
  3. [§Experiments, 'Q1: Boundary Adaptation and Stratification' and Conclusion] Table 1's zero observed harm is demonstrated on tasks generated from the same controlled boundary-family construction protocol used to build and validate the skills; Table 2B shows the same limitation, because the nonlinear boundary is 'unsupported' only until evolution is run on tasks from that same nonlinear family. The conclusion acknowledges that uncovered boundary families may still cause failure or harm, but the abstract and Table 1 present zero harm without this qualification. The claims should be restated as zero observed harm on the tested boundary families, not as a general runtime guarantee.
  4. [§Methodology, 'Validation-Gated Skill Evolution', Eqs. (8)-(10), and Table 1] Because T_sel and T_guard are used to decide promotion and the held-out evaluation tasks are drawn from the same boundary-family suite (new tasks, mixed boundaries, generator/severity transfer), the reported gains partly reflect model selection on tasks very similar to the selection tasks. The paper should report the number and construction of tasks in T_sel and T_guard, the performance gap between selection tasks and held-out tasks, and a clear statement of what generalization claim the experiments actually support.
minor comments (5)
  1. [Figure 3] The two panels are labeled 'A' and 'B' but their y-axes and exact quantities are not described in the caption; please add axis labels and a legend so the reader can see whether the plot shows AUC improvement, fallback rate, or both.
  2. [Table 3] The header 'Model Val. Harm Decision' is ambiguous; clarify that 'Valid' refers to schema-valid edits and 'Harm' to the guard-task harm rate, and make the column alignment consistent across all rows.
  3. [§Methodology, Eqs. (2)-(3)] The task distribution D in Eq. (2) and the guard distribution D_guard in Eq. (3) are not precisely defined; please state whether D_guard is a subset of D and how both relate to the controlled boundary suites.
  4. [References] Several references are formatted inconsistently (e.g., 'RUBIN 1976' appears in all caps, and some 2026 arXiv preprints lack archive identifiers); the reference list should be normalized to the journal style.
  5. [Abstract and Introduction] The text contains run-together words such as 'andmodeldevelopment' and 'Tothisend' in the abstract and introduction; these should be fixed in the camera-ready version.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: final-heldout evaluations and runtime evidence are separated from the promotion gate.

full rationale

SkillTFM's load-bearing mechanisms are Eq. (5) evidence extraction and Eq. (10) promotion gating. Eq. (5) computes z_T from the current task and base model without using test labels, and the paper states that 'Runtime evidence does not use current test labels or final-heldout feedback.' Eq. (10) gates candidate edits using promotion-selection tasks T_sel and guard tasks T_guard, but the paper explicitly reserves final-heldout tasks for reporting: 'final-heldout tasks are used only for final reporting.' The AUC gains and zero-harm results in Table 1 and Table 2B are reported on held-out boundary evaluations, not on the T_sel/T_guard sets used for promotion, so they are empirical results of the certificate and evolution process rather than direct restatements of the H_sel <= alpha constraint. The nonlinear skill in Panel B is promoted only after G_sel and H_sel checks, and the reported AUC improvement to 0.898 is on the final-heldout nonlinear boundary; this is standard selection-plus-heldout evaluation, not equivalence. The paper's stated limitation that 'uncovered boundary families may still cause failure or harm, and learning such skills requires additional validation data before deployment' is an external-validity caveat, not a circularity. Self-citations appear only in related work or as one of several base models and do not carry the derivation. The underspecification of r_sup, r_risk, and certificate thresholds is a rigor or evidence concern, not a circularity concern.

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

The central claim rests on a substantial number of hand-chosen thresholds and domain assumptions about the transferability of evidence-to-repair mappings. The promotion gate's epsilon, alpha, and delta are never quantified, and the runtime certificate's evidence and risk checks are left as qualitative criteria. The method also assumes that observable boundary evidence generalizes from development to deployment tasks, which is the main empirical risk.

free parameters (5)
  • epsilon (promotion improvement threshold)
    Gating in Eq. 10 requires G_sel(ΔS) >= epsilon; the value is never reported, so the reported improvements are conditional on an undisclosed threshold.
  • alpha (allowed harm rate)
    Eqs. 3, 9, and 10 constrain harm rates below alpha; without a reported alpha, 'zero observed harm' is not a falsifiable quantity.
  • delta (tolerated degradation margin)
    Harm is defined as utility below base by more than delta; delta is never stated, so the harm rate depends on an unknown margin.
  • runtime evidence sufficiency and risk thresholds
    The runtime certificate's checks ('evidence is sufficient', 'risk is acceptable') are not quantified; these thresholds determine when repairs are executed.
  • ranking priority rules and cost weights
    Eq. 7 defines a lexicographic ranking over support, history, risk, memory, and cost, but the priority order and cost(o) are not specified.
assumptions (4)
  • domain assumption Boundary failures are detectable from runtime evidence without test labels
    SkillTFM's gate relies on z_T = E(T, M) (Eq. 5) capturing repairable failure signals from covariates, predictions, and probes only; the paper explicitly states runtime evidence does not use test labels.
  • domain assumption Repair operators generalize within an evidence region
    Skills such as drop features, reweight samples, and output calibration are assumed to transfer across tasks sharing similar evidence; this is required for the skill bank to work beyond the exact promotion tasks.
  • domain assumption Promotion and guard task sets are representative of deployment
    The skill state is selected on T_sel and T_guard; the evaluation assumes these distributions match the held-out boundary and real-world conditions, which is plausible for the synthetic suite but unproven for electricity.
  • standard math Standard probability and decision-theoretic definitions
    The objective Eq. (2)-(3) uses standard expectation and probability notation; no nonstandard mathematical results are invoked.
invented entities (1)
  • External skill state S (evidence rules E, skill bank O, ranking R, guards G, fallback F, memory B, principles P)
    purpose: Serves as the training-free adaptation layer that decides between certified repair and fallback and accumulates validated skills.
    S is a software construct defined by the paper; it exists only inside SkillTFM and has no falsifiable handle outside the method itself. The paper provides no independent test that would confirm or refute the existence of such a state beyond its own AUC measurements.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SkillTFM: Gated Skill Evolution for Training-Free Adaptation of Tabular Foundation Models." pith.science (2026). https://pith.science/paper/WFQA7QBW

@misc{pith2026260806137,
  author       = {Pith},
  title        = {Pith review of: SkillTFM: Gated Skill Evolution for Training-Free Adaptation of Tabular Foundation Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WFQA7QBW}},
  note         = {Machine review of arXiv:2608.06137}
}
read the original abstract

Tabular data are ubiquitous in real-world applications and are crucial for data-driven prediction and decision-making across science, industry, finance, healthcare, and public services. Tabular foundation models (TFMs) have emerged as a promising paradigm for general-purpose tabular learning, offering reusable predictors across diverse datasets and substantially reducing the need for task-specific training, tuning, and model development. However, their practical deployment remains constrained by distribution shifts, heterogeneous feature semantics, and task-specific patterns that are difficult to capture without costly fine-tuning or additional labeled data. To this end, we propose SkillTFM, a training-free system that shifts TFM adaptation from parameter updates to the gated evolution of agentic skills. The core of SkillTFM is a verifiable and extensible skill bank that couples boundary evidence identification with gated skill evolution: the former characterizes task structure and base-model failure patterns, whereas the latter retrieves and extends reusable skills subject to explicit validation. Across simulated boundary settings and real-world electricity-price forecasting, SkillTFM improves AUC by 0.128--0.142, raises nonlinear-boundary AUC from 0.699 to 0.898. Furthermore, experiments across TFM backbones demonstrate the effectiveness and generality of SkillTFM.

Figures

Figures reproduced from arXiv: 2608.06137 by the authors.

Figure 1
Figure 1. Selective repair in electricity price forecasting windows from two cities. The annotated regions show how SkillTFM [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of SkillTFM. The lower panels show evidence-conditioned boundary expansion, where task evidence [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Boundary-stratified behavior under mixed [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Capability extension across frozen foundation models with a shared external skill state. Squares denote frozen-model [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 19 canonical work pages

  1. [1]

    Transformers Can Do Bayesian Inference , journal =

    Samuel M. Transformers Can Do Bayesian Inference , journal =. 2021 , url =. 2112.10510 , timestamp =

    arXiv 2021

  2. [2]

    2023 , eprint=

    TabPFN: A Transformer That Solves Small Tabular Classification Problems in a Second , author=. 2023 , eprint=

    2023

  3. [3]

    Proceedings of the 40th International Conference on Machine Learning , pages =

    Statistical Foundations of Prior-Data Fitted Networks , author =. Proceedings of the 40th International Conference on Machine Learning , pages =. 2023 , editor =

    2023

  4. [4]

    Nature , year =

    Hollmann, Noah and Müller, Samuel and Purucker, Lennart and Krishnakumar, Arjun and Körfer, Max and Hoo, Shi Bin and Schirrmeister, Robin Tibor and Hutter, Frank , title =. Nature , year =. doi:10.1038/s41586-024-08328-6 , url =

    doi:10.1038/s41586-024-08328-6

  5. [5]

    2025 , eprint=

    TabICL: A Tabular Foundation Model for In-Context Learning on Large Data , author=. 2025 , eprint=

    2025

  6. [6]

    2026 , eprint=

    TabDPT: Scaling Tabular Foundation Models on Real Data , author=. 2026 , eprint=

    2026

  7. [7]

    2025 , eprint=

    LimiX: Unleashing Structured-Data Modeling Capability for Generalist Intelligence , author=. 2025 , eprint=

    work page 2025

  8. [8]

    2025 , eprint=

    Real-TabPFN: Improving Tabular Foundation Models via Continued Pre-training With Real-World Data , author=. 2025 , eprint=

    2025

Show all 59 references
  1. [9]

    Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages =

    Chen, Tianqi and Guestrin, Carlos , title =. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages =. 2016 , isbn =. doi:10.1145/2939672.2939785 , abstract =

    2016

  2. [10]

    LightGBM: A Highly Efficient Gradient Boosting Decision Tree , url =

    Ke, Guolin and Meng, Qi and Finley, Thomas and Wang, Taifeng and Chen, Wei and Ma, Weidong and Ye, Qiwei and Liu, Tie-Yan , booktitle =. LightGBM: A Highly Efficient Gradient Boosting Decision Tree , url =

  3. [11]

    CatBoost: unbiased boosting with categorical features , url =

    Prokhorenkova, Liudmila and Gusev, Gleb and Vorobev, Aleksandr and Dorogush, Anna Veronika and Gulin, Andrey , booktitle =. CatBoost: unbiased boosting with categorical features , url =

  4. [12]

    Why do tree-based models still outperform deep learning on typical tabular data? , url =

    Grinsztajn, Leo and Oyallon, Edouard and Varoquaux, Gael , booktitle =. Why do tree-based models still outperform deep learning on typical tabular data? , url =

  5. [13]

    Tabular data: Deep learning is not all you need , journal =

    Ravid Shwartz-Ziv and Amitai Armon , keywords =. Tabular data: Deep learning is not all you need , journal =. 2022 , issn =. doi:https://doi.org/10.1016/j.inffus.2021.11.011 , url =

    2022 doi

  6. [14]

    and Leyton-Brown, Kevin , title =

    Thornton, Chris and Hutter, Frank and Hoos, Holger H. and Leyton-Brown, Kevin , title =. Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , pages =. 2013 , isbn =. doi:10.1145/2487575.2487629 , abstract =

    2013

  7. [15]

    Efficient and Robust Automated Machine Learning , url =

    Feurer, Matthias and Klein, Aaron and Eggensperger, Katharina and Springenberg, Jost and Blum, Manuel and Hutter, Frank , booktitle =. Efficient and Robust Automated Machine Learning , url =

  8. [16]

    2020 , eprint=

    AutoGluon-Tabular: Robust and Accurate AutoML for Structured Data , author=. 2020 , eprint=

    2020

  9. [17]

    , title =

    Geirhos, Robert and Jacobsen, Jörn-Henrik and Michaelis, Claudio and Zemel, Richard and Brendel, Wieland and Bethge, Matthias and Wichmann, Felix A. , title =. Nature Machine Intelligence , year =. doi:10.1038/s42256-020-00257-z , url =

    doi

  10. [18]

    Proceedings of the 38th International Conference on Machine Learning , pages =

    WILDS: A Benchmark of in-the-Wild Distribution Shifts , author =. Proceedings of the 38th International Conference on Machine Learning , pages =. 2021 , editor =

    2021

  11. [19]

    Covariate Shift Adaptation by Importance Weighted Cross Validation , journal =

    Masashi Sugiyama and Matthias Krauledat and Klaus-Robert M. Covariate Shift Adaptation by Importance Weighted Cross Validation , journal =. 2007 , volume =

    2007

  12. [20]

    , title =

    Guo, Chuan and Pleiss, Geoff and Sun, Yu and Weinberger, Kilian Q. , title =. Proceedings of the 34th International Conference on Machine Learning - Volume 70 , pages =. 2017 , publisher =

    2017

  13. [21]

    , journal=

    He, Haibo and Garcia, Edwardo A. , journal=. Learning from Imbalanced Data , year=

  14. [22]

    Journal of Big Data , year =

    Emmanuel, Tlamelo and Maupong, Thabiso and Mpoeleng, Dimane and Semong, Thabo and Mphago, Banyatsang and Tabona, Oteng , title =. Journal of Big Data , year =. doi:10.1186/s40537-021-00516-9 , url =

    doi

  15. [23]

    and Verleysen, M

    Frénay, B. and Verleysen, M. , title =. IEEE Transactions on Neural Networks and Learning Systems , year =. doi:10.1109/TNNLS.2013.2292894 , url =

    2013

  16. [24]

    Learning From Noisy Labels With Deep Neural Networks: A Survey

    Hwanjun Song and Minseok Kim and Dongmin Park and Yooju Shin and Lee, \ Jae Gil\. Learning From Noisy Labels With Deep Neural Networks: A Survey. IEEE Transactions on Neural Networks and Learning Systems. 2023. doi:10.1109/TNNLS.2022.3152527

    2023

  17. [25]

    and Raeder, Troy and Alaiz-Rodr\'

    Moreno-Torres, Jose G. and Raeder, Troy and Alaiz-Rodr\'. A unifying view on dataset shift in classification , year =. Pattern Recogn. , month = jan, pages =. doi:10.1016/j.patcog.2011.06.019 , abstract =

    2011 doi

  18. [26]

    , title =

    RUBIN, DONALD B. , title =. Biometrika , volume =. 1976 , month =. doi:10.1093/biomet/63.3.581 , url =

    1976 doi

  19. [27]

    and Nowozin, Sebastian and Dillon, Joshua and Lakshminarayanan, Balaji and Snoek, Jasper , booktitle =

    Ovadia, Yaniv and Fertig, Emily and Ren, Jie and Nado, Zachary and Sculley, D. and Nowozin, Sebastian and Dillon, Joshua and Lakshminarayanan, Balaji and Snoek, Jasper , booktitle =. Can you trust your model s uncertainty? Evaluating predictive uncertainty under dataset shift , url =

  20. [28]

    2018 , eprint=

    A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks , author=. 2018 , eprint=

    2018

  21. [29]

    Journal of Machine Learning Research , year =

    Ran El-Yaniv and Yair Wiener , title =. Journal of Machine Learning Research , year =

  22. [30]

    2019 , editor =

    Geifman, Yonatan and El-Yaniv, Ran , booktitle =. 2019 , editor =

    2019

  23. [31]

    2022 , eprint=

    A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantification , author=. 2022 , eprint=

    2022

  24. [32]

    2026 , eprint=

    SkillOpt: Executive Strategy for Self-Evolving Agent Skills , author=. 2026 , eprint=

    2026

  25. [33]

    2026 , eprint=

    SoK: Agentic Skills -- Beyond Tool Use in LLM Agents , author=. 2026 , eprint=

    2026

  26. [34]

    2026 , eprint=

    SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks , author=. 2026 , eprint=

    2026

  27. [35]

    2026 , eprint=

    SkillOS: Learning Skill Curation for Self-Evolving Agents , author=. 2026 , eprint=

    2026

  28. [36]

    2026 , eprint=

    CoEvoSkills: Self-Evolving Agent Skills via Co-Evolutionary Verification , author=. 2026 , eprint=

    2026

  29. [37]

    2026 , eprint=

    SKILLFOUNDRY: Building Self-Evolving Agent Skill Libraries from Heterogeneous Scientific Resources , author=. 2026 , eprint=

    2026

  30. [38]

    Proceedings of the 49th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =

    Liu, Xingyan and Luo, Xiyue and Li, Linyu and Huang, Ganghong and Liu, Jianfeng and Qiao, Honglin , title =. Proceedings of the 49th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages =. 2026 , isbn =. doi:10.1145/3805712.3808466 , ...

    2026

  31. [39]

    2026 , eprint=

    MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agents , author=. 2026 , eprint=

    2026

  32. [40]

    2026 , eprint=

    Agent Skills: A Data-Driven Analysis of Claude Skills for Extending Large Language Model Functionality , author=. 2026 , eprint=

    2026

  33. [41]

    2026 , eprint=

    A Comprehensive Survey on Agent Skills: Taxonomy, Techniques, and Applications , author=. 2026 , eprint=

    2026

  34. [42]

    2023 , eprint=

    Voyager: An Open-Ended Embodied Agent with Large Language Models , author=. 2023 , eprint=

    2023

  35. [43]

    2023 , eprint=

    Reflexion: Language Agents with Verbal Reinforcement Learning , author=. 2023 , eprint=

    2023

  36. [44]

    2024 , eprint=

    ExpeL: LLM Agents Are Experiential Learners , author=. 2024 , eprint=

    2024

  37. [45]

    2023 , eprint=

    ReAct: Synergizing Reasoning and Acting in Language Models , author=. 2023 , eprint=

    2023

  38. [46]

    2023 , eprint=

    Toolformer: Language Models Can Teach Themselves to Use Tools , author=. 2023 , eprint=

    2023

  39. [47]

    2023 , eprint=

    DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines , author=. 2023 , eprint=

    2023

  40. [48]

    Differentiation

    TextGrad: Automatic "Differentiation" via Text , author=. 2024 , eprint=

    2024

  41. [49]

    2026 , eprint=

    GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning , author=. 2026 , eprint=

    2026

  42. [50]

    2024 , eprint=

    Large Language Models as Optimizers , author=. 2024 , eprint=

    2024

  43. [51]

    2026 , publisher =

    Zhengkang Guan and Yikang Chen and Haoyuan Qian and Kairong Han and Peng Cui and Fei Wu and Kun Kuang , booktitle =. 2026 , publisher =

    2026

  44. [52]

    2606.21212 , archivePrefix=

    Zhengkang Guan and Yikang Chen and Yi He and Yunze Tong and Zijing Hu and Haoyuan Qian and Fei Wu and Kun Kuang , year=. 2606.21212 , archivePrefix=

    arXiv

  45. [53]

    2607.11510 , archivePrefix=

    Yikang Chen and Zhengkang Guan and Haoyuan Qian and Peng Cui and Yi Yang and Kun Kuang , year=. 2607.11510 , archivePrefix=

    arXiv

  46. [54]

    2026 , eprint=

    LimiX-2M: Mitigating Low-Rank Collapse and Attention Bottlenecks in Tabular Foundation Models , author=. 2026 , eprint=

    2026

  47. [55]

    2026 , howpublished =

    2026

  48. [56]

    2025 , howpublished =

    2025

  49. [57]

    2025 , eprint=

    MotherNet: Fast Training and Inference via Hyper-Network Transformers , author=. 2025 , eprint=

    2025

  50. [58]

    2024 , eprint=

    CARTE: Pretraining and Transfer for Tabular Learning , author=. 2024 , eprint=

    2024

  51. [59]

    2026 , eprint=

    TabPFN-2.5: Advancing the State of the Art in Tabular Foundation Models , author=. 2026 , eprint=

    2026

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.