Pith. sign in

REVIEW 4 major objections 6 minor 25 references

M-FISHER couples martingale shift detection with Fisher-preconditioned prompt updates to deliver anytime-valid false-alarm control and locally optimal adaptation for streaming vision-language models.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

M-FISHER combines an anytime-valid martingale shift detector with Fisher/natural-gradient prompt updates for streaming test-time adaptation of CLIP, with modest empirical gains and largely standard theory.

T0 review reviewed 2026-08-04 challenge →

load-bearing objection Reasonable system paper, but the anytime-valid guarantee is open-loop only – adaptation breaks it. the 4 major comments →

arxiv 2510.03839 v2 pith:L2K5EOBW submitted 2025-10-04 cs.LG stat.ML

Technical note on Sequential Test-Time Adaptation via Martingale-Driven Fisher Prompting

Behraj Khan , Tahir Qasim Syed This is my paper
classification cs.LG stat.ML
keywords sequential test-time adaptationmartingale detectione-processVille's inequalityFisher informationnatural gradientprompt tuningvision-language models
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 proposes M-FISHER, a sequential test-time adaptation framework that answers two coupled questions: when to adapt and how to adapt. For detection, it builds an exponential martingale from non-conformity scores, uses a bootstrap-corrected log-moment generating function, and applies Ville's inequality to guarantee a time-uniform false-alarm probability of at most α + 1/τ at any stopping time. For adaptation, it updates prompt embeddings via Fisher-preconditioned natural gradients, which the paper shows are locally KL-optimal and invariant to reparameterization. If correct, M-FISHER provides finite-sample, anytime-valid shift detection with a detection delay that scales as log(τ)/Γ, where Γ is the post-shift information growth rate, while keeping prompt updates stable and calibrated. A sympathetic reader would care because this unifies statistical sequential testing with geometric adaptation, potentially making test-time adaptation safer and more principled in safety-critical streaming applications.

Core claim

The paper's central claim is that sequential distribution shift detection and prompt adaptation can be unified in a theoretically grounded way. Concretely, the corrected process M_t = ∏ exp(λ(S_i − μ̂) − ψ̄(λ)) is shown to be a supermartingale conditional on a high-confidence bootstrap upper bound on the log-moment generating function, so Ville's inequality yields P(sup_t M_t ≥ τ) ≤ α + 1/τ unconditionally. Under sustained shift, the expected detection delay is bounded by log(τ)/Γ, where Γ = sup_λ {λ(E_{P1}[S] − μ_S) − ψ(λ)} is the post-shift exponential growth rate, linking detection speed to distributional divergence. For adaptation, a Fisher-preconditioned step P ← P − η(F_P + γI)^{-1}∇L

What carries the argument

The central objects are (i) an exponential supermartingale (e-process) built from non-conformity scores S_t = KL(p_θ(y|x_t) ‖ Uniform) + α·Mahalanobis distance, with a bootstrap-corrected log-MGF ψ̄(λ) that restores the supermartingale property under finite calibration data; and (ii) the Fisher Information Matrix F_P of the prompt parameters, which preconditions gradient updates to implement natural gradient descent in the Fisher–Rao metric. Ville's inequality converts the supermartingale property into a time-uniform false-alarm bound valid at arbitrary stopping times, while the Fisher geometry ensures that each update is locally KL-optimal and stable under reparameterization. The combinatio

Load-bearing premise

The load-bearing premise is that the held-out calibration set accurately represents the null (no-shift) score distribution for the entire stream, so that the bootstrap-derived high-confidence bound on ψ(λ) is a genuine upper bound; if the calibration set is non-representative or the bootstrap undercovers, the claimed false-alarm guarantee α + 1/τ can fail.

What would settle it

Run the detector on a stream where the calibration set is drawn from a distribution that differs only slightly from the test stream's null distribution (e.g., a different domain or time period). If the empirical false-alarm rate consistently exceeds α + 1/τ, or if the bootstrap quantile fails to cover the true ψ(λ) in a simulation study, the central validity claim is refuted. A second decisive test: measure expected detection delay on a sustained shift and check whether it scales as log(τ)/Γ across multiple thresholds τ; if the scaling breaks or the constant is far off for non-IID streams, the

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the false-alarm bound α + 1/τ holds, practitioners can set τ to target a desired false-alarm rate and trust the detector at any stopping time, enabling safe automation in streaming decision systems.
  • The detection-delay scaling log(τ)/Γ quantifies how shift magnitude (via Γ) trades off against detection speed, giving a principled way to set thresholds based on the expected cost of delay versus false alarms.
  • Fisher-preconditioned updates mean that after a detected shift, prompt adaptation makes the locally smallest KL step that improves the detection loss, reducing the risk of overfitting to transient noise compared to unconstrained gradient updates.
  • The anytime-valid nature of the detector means the same guarantee holds even if the stream length is unknown or unbounded, which is natural for online deployment.
  • The ablation results, if reproducible, imply that both the martingale trigger and Fisher preconditioning contribute complementary gains, suggesting that trigger-then-adapt is a better design than continuous or fixed-interval adaptation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The unconditional bound α + 1/τ hinges on the bootstrap quantile being a genuine 1−α upper confidence bound on ψ(λ); this is asserted but not proven, so the practical guarantee could degrade if the bootstrap undercovers or the calibration set is not representative of the null distribution.
  • The Γ-based delay bound is asymptotic and assumes IID or well-mixing score processes; for dependent or adversarial streams, the log(τ)/Γ scaling may only hold approximately, and empirical delay could be much larger—so the bound should be treated as a design heuristic outside those regimes.
  • Because the non-conformity score mixes confidence (KL from uniform) with feature-space distance, Γ implicitly depends on both predictive uncertainty and feature drift; this suggests the method could be extended to detect label shift or subpopulation shifts by redefining the score, though the paper does not test that.
  • The reported empirical gains are modest (1–3% accuracy, 0.5-sample delay improvement), but the theoretical guarantees—if valid—justify using the method in settings where false alarms are costly, even when average accuracy improvements are small.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. This paper proposes M-FISHER, a sequential test-time adaptation method for vision-language models that couples martingale-based shift detection with Fisher-preconditioned prompt updates. The detection component forms an exponential process from non-conformity scores (Eq. 3/4), declares a shift when the process exceeds a threshold τ, and claims time-uniform false-alarm control via Ville's inequality (Eq. 10). The adaptation component updates prompt parameters with a damped diagonal Fisher preconditioner (Eq. 6–7), argued to be a natural-gradient step with local KL-optimality and reparameterization invariance (Theorem A.1). A detection-delay bound of order log(τ)/Γ is claimed under sustained shifts (Eq. 11–12). Experiments on ImageNet-C, ImageNet-R, and Office-Home report accuracy, calibration error, detection delay, and false-alarm rate, with ablations separating the contribution of the martingale trigger and the Fisher preconditioner.

Significance. If the theoretical claims held as stated, M-FISHER would be a useful template for principled TTA: the use of e-processes/martingales for anytime-valid detection is well motivated, the information-geometric perspective on prompt updates is standard, and the paper explicitly releases code and includes ablations. These are genuine strengths. However, the central anytime-valid guarantee is not established for the closed-loop algorithm as deployed, because prompt adaptation after a detection changes the score distribution and invalidates the supermartingale argument. The bootstrap finite-sample correction is asserted rather than proved, and the detection-delay claim relies on classical results whose hypotheses are not verified for the exponential e-process with plug-in estimates and post-adaptation score dynamics. These are load-bearing gaps, not presentation issues.

major comments (4)
  1. [§4.1 and Algorithm 1, Eq. (10)] The supermartingale property and Ville bound (Eq. 10) hold only while the scores S_t are drawn from the fixed null distribution used to calibrate μhat and ψbar. In Algorithm 1, whenever M_t ≥ τ, the prompt P is updated via Eq. (7). After that first update — including after a false alarm — the score distribution changes, so the increments exp(λ(S_i−μhat)−ψbar(λ)) are no longer conditionally mean ≤ 1 under the null. The time-uniform guarantee therefore does not apply to the full M-FISHER process as deployed and as evaluated in Table 1. The paper should either state that the guarantee holds only up to the first adaptation, or prove a closed-loop supermartingale property under the adaptation rule, or present Table 1 as a validation of the detection module with adaptation disabled.
  2. [§3.2 and Proposition A.1] The 'finite-sample correction' is not proven. The proposition assumes P(ψ(λ) ≤ ψbar(λ)) ≥ 1−α and derives the unconditional bound α + 1/τ, but the main text does not prove that the empirical (1−α)-quantile of the bootstrap MGF provides such coverage, let alone uniformly over λ and over the stream. Bootstrap quantiles are first-order asymptotically valid under regularity conditions, not exact finite-sample confidence bounds. Moreover, μhat is also estimated on the same calibration set and appears inside the exponent, and the bootstrap procedure described in §3.2 does not account for this extra estimation step. Without a proof or a precise asymptotic statement, the advertised finite-sample false-alarm control is not established.
  3. [§4.2, Eqs. (11)–(12)] The detection-delay bound is asserted, not derived for the actual detector. Lorden's classical result applies to likelihood-ratio procedures with known pre-change and post-change distributions. Eq. (11) is invoked as 'classical', but the extension to exponential e-processes with plug-in μhat and ψbar, unknown P1, and a stopping rule based on M_t is not proved in the main text. Theorem A.2 gives an asymptotic IID statement under a fixed prompt, but Algorithm 1 changes the prompt upon detection, so E_{P1}[(T(τ)−t*)+] ≲ log(τ)/Γ is not a property of the closed-loop method. If the paper intends this as an analysis of the detection module in isolation, that limitation should be explicitly stated.
  4. [§4.3 and Theorem A.1] Theorem A.1 is stated and proved for the exact Fisher Information Matrix F_P, but Algorithm 1 and Eq. (6) use a damped diagonal estimator. The steepest-descent interpretation and reparameterization invariance in Theorem A.1 do not generally survive diagonalization or damping. The paper should either justify that the approximate update retains these properties under the stated assumptions, or weaken the 'geometrically stable' claim to hold only for the exact natural-gradient update.
minor comments (6)
  1. [Throughout] There are several missing references and placeholders: 'Ville's inequality [?]' after Eq. (10), '?Liang et al., 2020' in the introduction, 'Figure??' before Figure 1, and 'see detail A.1' with no clear pointer. These should be cleaned up.
  2. [§3.2, Eq. (5)] The notation is inconsistent: Eq. (5) writes conditioning on C, but the text says 'conditional on the calibration event {ψ(λ) ≤ ψbar(λ)}'. Proposition A.1 conditions on the event. Please make the conditioning object explicit.
  3. [Algorithm 1] Algorithm 1 line 1 says 'Compute μS and ψ(λ)', while the corrected process in §3.2 uses ψbar(λ). Line 5 says 'Update M_t via Eq. 3', but Eq. (3) uses ψ, not ψbar. The algorithm and the main-text equations should be aligned.
  4. [§5.1 and Table 2] DomainNet is listed in §5.1 and mentioned in the conclusion, but no DomainNet results appear in Table 2. Either report the results or remove the claim of evaluation on DomainNet.
  5. [§6.1, Table 1] The detection statistics are averaged over only three random shift orderings. The paper itself notes the need for more runs; please provide per-shift-type breakdowns and confidence intervals or additional seeds before making strong comparative claims.
  6. [Typos] There are typos such as 'evlaute', 'perfroamnce', and 'benchmarking' in §5.1, and 'more' fragment in the reader's summary is not the issue; in the manuscript, please run a careful proofread.

Circularity Check

0 steps flagged

No significant circularity: the derivation chain rests on external theorems (Ville, Lorden, Amari) and calibration is used as an explicit conditioning assumption, not as a fitted prediction.

full rationale

Walking the derivation chain: the false-alarm claim (Eqs. 4-5 and Prop. A.1) is a conditional supermartingale/union-bound argument. The calibration estimates μ̂ and ψ̄ are inputs, but the bound α + 1/τ is not obtained by fitting the bound to alarm data; it is Ville's inequality applied to a process whose increments have conditional mean ≤1 under the calibration event. The bootstrap coverage condition is assumed rather than proved, which is a missing technical justification, not a circular reduction. The delay bound (Eqs. 11-12) imports Lorden's theorem and defines Γ as the per-step exponential growth rate; log τ/Γ follows from the e-process definition and classical change-detection theory, not from the paper's own fitted constants. The Fisher-prompt update is justified by Amari's natural gradient theory, an external result. The self-citations (Khan et al. 2024/2025) are contextual/baseline references and are not load-bearing. The empirical false-alarm rate at τ=100 is a check of the bound, not a reproduction of fitted values. A more serious concern is that after the first prompt update the score distribution changes and the fixed-null supermartingale may no longer hold for the closed-loop system; this is a validity/correctness gap, and the paper's own Limitations (A.2) concede calibration and sustained-shift assumptions, but it is not a case of a prediction being equivalent to its inputs by construction.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 0 invented entities

The central claims rest on classical martingale/change-point/natural-gradient theorems (standard math) plus three load-bearing domain assumptions that the paper does not fully verify: calibration-set validity and bootstrap MGF upper bound, mixing/IID score processes, and equivalence of the damped diagonal Fisher to the full FIM for the natural-gradient guarantees. The free parameters are mostly hyperparameters; none is fit to reproduce the theoretical claims, but λ and α are unreported.

free parameters (6)
  • λ (martingale exponential parameter) = not reported
    Appears in Eq. 3/4 and defines the e-process and Γ; no selection rule or value is given, yet the false-alarm and delay guarantees depend on it.
  • α (non-conformity score scaling) = not reported
    Scales the Mahalanobis term in Eq. 1; no value or tuning protocol is reported.
  • η (prompt learning rate) = selected from {1e-5, 5e-5, 1e-4} via calibration-set grid search
    Hyperparameter chosen on calibration data, affecting adaptation accuracy and stability.
  • τ (martingale threshold) = 100
    Set to target ≤1% false alarms via τ = 1/δ, but the stated unconditional bound is α + 1/τ, so the δ/α relationship is not fully specified.
  • λdamp (Fisher damping) = 1e-4
    Damping constant for diagonal Fisher inverse; controls step size and stability.
  • k (prompt length) = 16
    Number of learnable prompt tokens; model capacity choice.
axioms (6)
  • standard math Ville's inequality for nonnegative supermartingales
    Used in Eq. 5/10 to claim P(sup Mt ≥ τ) ≤ 1/τ; this is a classical result, but the paper references it with a broken citation marker.
  • domain assumption Lorden's change-point delay bound applies to the exponential e-process with Γ growth rate
    §4.2 asserts E_P1[(T−t*)+] ≲ log τ / KL and extends to exp e-process 'same scaling holds' with Γ; requires pre/post score distributions satisfying detectability/regularity and e-process lower-bounding likelihood ratios, which is only sketched.
  • ad hoc to paper Calibration set C is independent of the test stream and representative of the null distribution; bootstrap quantile is a valid 1−α upper confidence bound for ψ(λ)
    §3.2 finite-sample correction replaces ψ with empirical (1−α)-quantile of bootstrap MGF estimates, but no theorem establishes P(ψ ≤ ψbar) ≥ 1−α; the paper's own limitations admit calibration set may be unavailable/nonstationary.
  • domain assumption Scores S_t are IID (or sufficiently mixing) under null and post-shift for delay bound
    Appendix A.1 Theorem A.2 and Prop A.2 assume IID or α-mixing with finite MGF; the main text uses these conditions implicitly without checking them on real benchmarks.
  • ad hoc to paper A damped diagonal Fisher estimator preserves natural-gradient optimality and reparameterization invariance (Theorem A.1)
    §3.3/§5.2 use F_P ≈ damped diagonal; the KL-optimality and invariance theorems require the full FIM, and Kunstner et al. (cited) document failures of empirical Fisher approximations.
  • standard math Amari natural gradient steepest-descent result
    Theorem A.1 restates the classical natural-gradient result; sound if full FIM is used and the model is differentiable.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Technical note on Sequential Test-Time Adaptation via Martingale-Driven Fisher Prompting." pith.science (2026). https://pith.science/paper/L2K5EOBW

@misc{pith2026251003839,
  author       = {Pith},
  title        = {Pith review of: Technical note on Sequential Test-Time Adaptation via Martingale-Driven Fisher Prompting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/L2K5EOBW}},
  note         = {Machine review of arXiv:2510.03839}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We present a theoretical framework for M-FISHER, a method for sequential distribution shift detection and stable adaptation in streaming data. For detection, we construct an exponential martingale from non-conformity scores and apply Ville's inequality to obtain time-uniform guarantees on false alarm control, ensuring statistical validity at any stopping time. Under sustained shifts, we further bound the expected detection delay as $\mathcal{O}(\log(1/\delta)/\Gamma)$, where $\Gamma$ reflects the post-shift information gain, thereby linking detection efficiency to distributional divergence. For adaptation, we show that Fisher-preconditioned updates of prompt parameters implement natural gradient descent on the distributional manifold, yielding locally optimal updates that minimize KL divergence while preserving stability and parameterization invariance. Together, these results establish M-FISHER as a principled approach for robust, anytime-valid detection and geometrically stable adaptation in sequential decision-making under covariate shift.

Figures

Figures reproduced from arXiv: 2510.03839 by Behraj Khan, Tahir Qasim Syed.

Figure 1
Figure 1. Figure 1: Sequential shift detection and adaptation in M-FISHER. The prediction distribution (blue) is monitored via [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Ablation analysis on ImageNet-C sequential shift. Accuracy (gray) and Expected Calibration Error (blue) are [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: t-SNE visualization of text prompt embeddings before and after M-FISHER adaptation on Office-Home [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

25 extracted references · 3 linked inside Pith

  1. [1]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pages 8748--8763. PMLR, 2021

    2021

  2. [2]

    Density ratio estimation in machine learning

    Masashi Sugiyama, Taiji Suzuki, and Takafumi Kanamori. Density ratio estimation in machine learning. Cambridge University Press, 2012

    2012

  3. [3]

    On calibration of modern neural networks

    Chuan Guo, Geoff Pleiss, Yu Sun, and Kilian Q Weinberger. On calibration of modern neural networks. In International conference on machine learning, pages 1321--1330. PMLR, 2017

    2017

  4. [4]

    Learning to prompt for vision-language models

    Kaiyang Zhou, Jingkang Yang, Chen Change Loy, and Ziwei Liu. Learning to prompt for vision-language models. International Journal of Computer Vision, 130 0 (9): 0 2337--2348, 2022

    2022

  5. [5]

    Prompt distribution learning

    Yuning Lu, Jianzhuang Liu, Yonggang Zhang, Yajing Liu, and Xinmei Tian. Prompt distribution learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5206--5215, 2022

    2022

  6. [6]

    Tent: Fully test-time adaptation by entropy minimization

    Dequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno Olshausen, and Trevor Darrell. Tent: Fully test-time adaptation by entropy minimization. arXiv preprint arXiv:2006.10726, 2020

    Pith/arXiv arXiv 2006

  7. [7]

    Do we really need to access the source data? source hypothesis transfer for unsupervised domain adaptation

    Jian Liang, Dapeng Hu, and Jiashi Feng. Do we really need to access the source data? source hypothesis transfer for unsupervised domain adaptation. In International conference on machine learning, pages 6028--6039. PMLR, 2020

    2020

  8. [8]

    Gradient matching for domain generalization

    Yuge Shi, Jeffrey Seely, Philip HS Torr, Narayanaswamy Siddharth, Awni Hannun, Nicolas Usunier, and Gabriel Synnaeve. Gradient matching for domain generalization. arXiv preprint arXiv:2104.09937, 2021

    Pith/arXiv arXiv 2021

  9. [9]

    Confidence-calibrated covariate shift correction for few-shot classification in vision-language models

    Behraj Khan, Rizwan Qureshi, Nouman Muhammad Durrani, and Tahir Qasim Syed. Confidence-calibrated covariate shift correction for few-shot classification in vision-language models. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 6511--6523, 2025

    2025

  10. [10]

    Causal covariate shift correction using fisher information penalty

    Behraj Khan, Behroz Mirza, and Tahir Syed. Causal covariate shift correction using fisher information penalty. In The Second Tiny Papers Track at ICLR 2024, 2024

    2024

  11. [11]

    Sequential nonparametric testing with the law of the iterated logarithm

    Akshay Balsubramani and Aaditya Ramdas. Sequential nonparametric testing with the law of the iterated logarithm. arXiv preprint arXiv:1506.03486, 2015

    Pith/arXiv arXiv 2015

  12. [12]

    Can autonomous vehicles identify, recover from, and adapt to distribution shifts? In International Conference on Machine Learning, pages 3145--3153

    Angelos Filos, Panagiotis Tigkas, Rowan McAllister, Nicholas Rhinehart, Sergey Levine, and Yarin Gal. Can autonomous vehicles identify, recover from, and adapt to distribution shifts? In International Conference on Machine Learning, pages 3145--3153. PMLR, 2020

    2020

  13. [13]

    Failing loudly: An empirical study of methods for detecting dataset shift

    Stephan Rabanser, Stephan G \"u nnemann, and Zachary Lipton. Failing loudly: An empirical study of methods for detecting dataset shift. Advances in Neural Information Processing Systems, 32, 2019

    2019

  14. [14]

    Covariate shift by kernel mean matching

    Arthur Gretton, Alex Smola, Jiayuan Huang, Marcel Schmittfull, Karsten Borgwardt, Bernhard Sch \"o lkopf, et al. Covariate shift by kernel mean matching. Dataset shift in machine learning, 3 0 (4): 0 5, 2009

    2009

  15. [15]

    Overcoming catastrophic forgetting in neural networks

    James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, et al. Overcoming catastrophic forgetting in neural networks. Proceedings of the national academy of sciences, 114 0 (13): 0 3521--3526, 2017

    2017

  16. [16]

    Natural gradient works efficiently in learning

    Shun-Ichi Amari. Natural gradient works efficiently in learning. Neural computation, 10 0 (2): 0 251--276, 1998

    1998

  17. [17]

    Limitations of the empirical fisher approximation for natural gradient descent

    Frederik Kunstner, Philipp Hennig, and Lukas Balles. Limitations of the empirical fisher approximation for natural gradient descent. Advances in neural information processing systems, 32, 2019

    2019

  18. [18]

    Procedures for reacting to a change in distribution

    Gary Lorden. Procedures for reacting to a change in distribution. The annals of mathematical statistics, pages 1897--1908, 1971

    1908

  19. [19]

    Probability with martingales

    David Williams. Probability with martingales. Cambridge university press, 1991

    1991

  20. [20]

    Jean ville, \'e tude critique de la notion de collectif

    Joseph L Doob. Jean ville, \'e tude critique de la notion de collectif. 1939

    1939

  21. [21]

    Time-uniform, nonparametric, nonasymptotic confidence sequences

    Steven R Howard, Aaditya Ramdas, Jon McAuliffe, and Jasjeet Sekhon. Time-uniform, nonparametric, nonasymptotic confidence sequences. The Annals of Statistics, 49 0 (2): 0 1055--1080, 2021

    2021

  22. [22]

    The many faces of robustness: A critical analysis of out-of-distribution generalization

    Dan Hendrycks, Steven Basart, Norman Mu, Saurav Kadavath, Frank Wang, Evan Dorundo, Rahul Desai, Tyler Zhu, Samyak Parajuli, Mike Guo, et al. The many faces of robustness: A critical analysis of out-of-distribution generalization. In Proceedings of the IEEE/CVF international conference on computer vision, pages 8340--8349, 2021

    2021

  23. [23]

    Deep hashing network for unsupervised domain adaptation

    Hemanth Venkateswara, Jose Eusebio, Shayok Chakraborty, and Sethuraman Panchanathan. Deep hashing network for unsupervised domain adaptation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5018--5027, 2017

    2017

  24. [24]

    Moment matching for multi-source domain adaptation

    Xingchao Peng, Qinxun Bai, Xide Xia, Zijun Huang, Kate Saenko, and Bo Wang. Moment matching for multi-source domain adaptation. In Proceedings of the IEEE/CVF international conference on computer vision, pages 1406--1415, 2019

    2019

  25. [25]

    Obtaining well calibrated probabilities using bayesian binning

    Mahdi Pakdaman Naeini, Gregory Cooper, and Milos Hauskrecht. Obtaining well calibrated probabilities using bayesian binning. In Proceedings of the AAAI conference on artificial intelligence, volume 29, 2015

    2015

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.