Pith. sign in

REVIEW 4 major objections 4 minor 53 references

ALSA: Anchors in Logit Space for Out-of-Distribution Accuracy Estimation

T0 review · 4 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A handful of learnable anchors in logit space can estimate a model's accuracy on unlabeled, shifted data without any target labels.

desk verdict ALSA is a practical, empirically strong accuracy estimator that is genuinely new in mechanism, but its theory is thin and its 'wide range' claim runs ahead of what the benchmarks actually test. read the letter →

arxiv 2508.19613 v1 pith:GCH6GTNW submitted 2025-08-27 cs.LG

Chenzhi Liu , Mahsa Baktashmotlagh , Yanran Tang , Zi Huang , Ruihong Qiu This is my paper · ORCID
classification cs.LG
keywords out-of-distributionaccuracyestimationlogitspacelearnableanchorsdistributionshiftunlabeledmodelevaluationsoftmaxinformationlossclassimbalance
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that a model's accuracy on an unlabeled, out-of-distribution dataset can be estimated by looking at where its logits fall in a low-dimensional band of logit space. The key claim is that logits carry more information about whether a prediction is correct than softmax scores do, because softmax throws away magnitude and class-specific structure. The proposed method, ALSA, places learnable reference points, called anchors, in logit space; each anchor exerts a distance-decaying influence, and the total influence is converted into a per-sample probability of correctness. Averaging those probabilities over the unlabeled target set yields an accuracy estimate. The paper reports consistent gains over softmax-based and similarity-based baselines across vision, language, and graph benchmarks under several kinds of distribution shift.

What carries the argument

The central object is a set of anchor triplets (position a, peak influence p, variance v) in logit space, combined through a Gaussian-like influence function that decays with cosine distance from the anchor. The total influence from all anchors is mapped by a sigmoid into a per-sample probability of correct prediction, and the average of these probabilities is the accuracy estimate. The band-width proposition justifies the geometric assumption that logits concentrate near a hyperplane, while the rectification threshold, derived from a confidence interval through the inverse error function, handles low-influence regions where the estimate would otherwise revert to an uninformative 50%.

What would settle it

Construct a target set by taking source validation logits and relabeling a contiguous region of logit space as incorrect while leaving the marginal logit distribution unchanged, so that p(correct|z) differs between source and target. ALSA, whose anchors were trained on the unaltered source labels, would then estimate accuracy by averaging correctness probabilities from the source surface, and the deviation between its estimate and the relabeled target accuracy would be roughly the probability mass of the relabeled region.

Watch

Extended reading notes

Core claim

The paper claims that a classifier's logits lie in a narrow band around a (c-1)-dimensional hyperplane perpendicular to the all-ones vector, and that within that band a logit's position carries reliable information about whether the prediction will be correct. ALSA formalizes this by initializing k learnable anchors in logit space, each defined by a position vector, a signed peak influence, and a variance controlling how far its influence reaches. The total influence from all anchors is passed through a sigmoid to give the probability that the corresponding sample is correctly predicted. The estimated accuracy on an unlabeled target set is the average of these per-sample probabilities, with

Load-bearing premise

The anchors are learned on the source validation set and applied to the target set, so ALSA works only if the probability of a correct prediction given a logit vector is approximately the same in source and target; the paper notes this can break down under extremely severe corruptions.

Editorial extensions

If this is right

  • If ALSA's central claim holds, deployed models can be monitored on unlabeled test data by averaging per-sample correctness probabilities derived from logit positions, without needing target labels.
  • Avoiding softmax compression preserves logit magnitude information, which should improve estimates in settings where class distributions are imbalanced and where accuracy differs across regions of logit space.
  • The same anchor mechanism transfers across vision, language, and graph models, suggesting a general-purpose estimator rather than a domain-specific similarity score.
  • Because inference scales linearly with the number of samples and anchors, the method could be used in production monitoring with modest compute overhead.
  • ALSA is reported to remain accurate even when true accuracy drops by over 30% under synthetic corruptions, and to keep mean absolute error below the levels where several baselines fail on at least one dataset.

Reading between the lines

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

  • Beyond the paper, the learned anchor surface could double as a principled confidence or selective-classification score, since it assigns a calibrated correctness probability to every logit region rather than a single global threshold.
  • The disclosed assumption that p(correct|z) aligns between source and target suggests ALSA should be stress-tested under explicit label shift, where the same logit region may have different correctness rates; the paper's benchmarks cover covariate-like shifts more than pure label shifts.
  • If anchor positions and peak values converge to the same distribution regardless of initialization, as the paper argues, then the anchor set is effectively approximating an underlying correctness surface, which would make ALSA a candidate building block for other logit-based analyses such as OOD detection.
  • The influence-based aggregation could be extended to produce per-class accuracy estimates, which the current average-based estimate does not directly expose, potentially aiding debugging of skewed classifiers.
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 / 4 minor

Summary. ALSA estimates the accuracy of a classifier on an unlabeled target dataset by placing k learnable anchors in logit space, each with a peak value and variance, and defining the per-sample correctness probability as the sigmoid of the sum of Gaussian-like or exponential influence functions. The anchors are trained on a labeled source validation set by binary cross-entropy (Eq. (9)); the target accuracy estimate is the average of these probabilities over target logits (Eq. (7)), with a rectification for low-influence regions. The method is evaluated on vision, language, and graph benchmarks—CIFAR-10/100C, CIFAR-10.1/10.2, ImageNetV2, MNIST-M, Office-31, Waterbirds/Amazon WILDS, and ogbn-arxiv—across multiple architectures, reporting MAE and correlation against baselines (AC, DoC, IM, ATC, COT, GNNEval). Code is released.

Significance. If the empirical results hold, ALSA is a practical, unified OOD accuracy estimator with a favorable accuracy/efficiency trade-off. The paper's strengths include a broad multi-modal evaluation, ablations of learnable components, sensitivity analyses for alpha and anchor count, and a complexity analysis; the release of code is a plus. The main limitation is that the method's core assumption—that p(correct|z) transfers from source to target—is only tested indirectly, and one proposition is not proven for trained networks. The contribution is significant for the accuracy-estimation subfield if that assumption and the reported results are made explicit and properly scoped.

major comments (4)
  1. [Section 3.3 / 3.4, Appendix I.3, H.2] The target estimator is a plug-in for E_{P_T}[p_S(correct|Z)]: anchors are fit on source validation via Eq. (9), and Eq. (7) evaluates that source-fitted function on target logits. The method is therefore unbiased only if p_T(correct|z)=p_S(correct|z) on the support of the target logits. The paper acknowledges this in Appendix I.3 but only as a caveat for extreme corruptions. None of the main benchmarks induces a genuine conditional shift: CIFAR-C, MNIST-M, and Office-31 are covariate shifts; WILDS changes group frequencies with a stable labeling function; the class-imbalance training in H.2 (1:3) is mild. Thus the claim of robustness 'across a wide range of distribution shifts' is under-tested. I recommend adding an experiment that changes the label prior or labeling function (e.g., label shift with rebalanced target classes, or a synthetic shift that changes class confusability) and sc
  2. [Section 2.2 / Appendix B] Proposition 1 derives a bandwidth bound for a randomly initialized linear layer with W ~ N(0, sigma_W), b=0, and Xavier initialization. The text then argues that training updates are confined to directions orthogonal to the all-ones vector, but no proof shows that the magnitude bound (Bandwidth < 4z sqrt(1/(m+c))) persists after training. Since ALSA is applied to trained networks, the proposition as stated does not cover the setting of interest. It should either be explicitly labeled as an initialization-only observation or supplemented by an empirical/theoretical argument for trained networks.
  3. [Section 3.3 vs Appendix E] There is an inconsistency in the rectification condition. Eq. (8) defines insufficient influence as |Infl(z_i, A)| < t, where Infl(z_i, A) is the total influence from all anchors, while Appendix E defines it as |Infl(z_i, (a_j,p_j,v_j))| < t for all j individually. These are different conditions and can lead to different predictions. In addition, Section 3.3 leaves t in terms of a per-anchor p, while Appendix E fixes p=6 to obtain a unified threshold; anchors with smaller learned p are then treated with the same threshold. The paper should state which condition is implemented and justify the unified p=6 threshold given that p is a learned parameter.
  4. [Appendix J / main experiments] The paper does not report the alpha value used for Tables 1 and 2. This hyperparameter controls when low-influence logits are replaced by 1/c and is shown in Appendix J to affect MAE substantially at extreme values. For reproducibility, the alpha values (and anchor counts) for each benchmark should be listed in the main text or in an experiment-setting table. The code release is helpful, but the paper itself should specify the default configuration.
minor comments (4)
  1. [Section 3.3 / 3.4] Eq. (6) defines an 'initial' accuracy estimate before rectification, and Eq. (7) the final one. Section 3.4's stopping criterion uses p_true from Eq. (6) but does not explicitly state whether p_rec from Eq. (7) is used during training. Please clarify.
  2. [Table 3, GIN row] ALSA obtains R^2 = -0.32 and Pearson r = 0.46 on GIN, which is lower than several baselines' correlation coefficients. The sentence 'consistently achieves high correlation scores across models' is too strong and should be qualified, or the authors should explain why R^2 is negative (e.g., small variance in true accuracy across test years).
  3. [Table 1, Office-31] Office-31 results have very large standard deviations (e.g., 15.29 for AC, 33.25 for ATC). Consider reporting per-domain results separately or discussing this instability, since the averaged MAE may be dominated by a few seeds/domains.
  4. [Figure 11] Figure 11 reports MSE (mean squared error) while Tables 1 and 2 use MAE. Please use one metric consistently to avoid confusion.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: source-fitted anchors are applied out-of-sample to unlabeled target logits; the disclosed conditional-alignment assumption is a limitation, not a circular step.

full rationale

The derivation is self-contained as an out-of-sample estimator. Anchors are learned on the labeled source validation set by minimizing the BCE loss in Eq. (9); the target accuracy estimate in Eq. (7) averages the sigmoid of summed anchor influences (Eqs. (1)-(4), with rectification in Eq. (8)) over target logits only. No target labels or target accuracy enter training or inference. The stopping rule in Eq. (10) compares predicted and true accuracy on the validation set, so it is a model-selection signal, not a constraint on the target estimate. Appendix I.3 discloses the method's reliance on approximate alignment of p(y|z) between source and target and notes it can weaken under extreme corruption; this is an honest scoping limitation, not a circular reduction. The only author self-citations ([45,46]) appear in a related-work enumeration and are not load-bearing for ALSA's claims. I find no step where a fitted parameter is renamed as a target prediction or where a prediction reduces by construction to its inputs.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the transfer of a correctness-likelihood model from source validation to target, on a specific functional form for influence, and on a set of per-dataset hyperparameters. No new physical or conceptual entities are postulated; anchors are learnable parameters of the model.

free parameters (4)
  • confidence interval alpha = not reported; appendix sweeps 0.7-0.999
    Controls the rectification threshold t (Eq. 8 and Appendix E). The value used to produce Tables 1-3 is not stated in the paper.
  • number of anchors k = 3000 (CIFAR10/100), 20000 (ImageNet), 5000 (Amazon-WILDS), 500 (Waterbirds), 3000 (MNIST), 300 (Office-31)
    Per-dataset hyperparameter listed in Table 6; accuracy estimates depend on k, though the paper shows robustness above a threshold.
  • peak value p = 6
    Chosen so sigmoid(6)=99.75%; used to derive the unified threshold t for rectification.
  • influence function choice = Gaussian-like vs exponential
    ALSA-G and ALSA-E are both reported; results differ notably (e.g., CIFAR-10.2 MAE 4.19 vs 0.61), so the functional form is a consequential modeling choice.
assumptions (5)
  • domain assumption Logits from trained classifiers lie near a hyperplane perpendicular to the all-ones vector (bandwidth bound).
    Proposition 1 is proved for weights initialized as iid Gaussians with independent embeddings, but is asserted to hold for trained networks in Section 2.2 and used to motivate the anchor design.
  • ad hoc to paper Correctness probability is a smooth function of logit position, well approximated by a sum of Gaussian-like influences.
    The influence function in Eq. 1 is a modeling choice; no proof is given that it can represent p(correct|z) in general.
  • domain assumption Approximate alignment of p(y|z) between source and target (covariate shift).
    Stated in Appendix I.3 as the method's underlying assumption; if violated, the learned anchors give biased estimates.
  • ad hoc to paper Cosine distance is an appropriate similarity measure in logit space.
    Eq. 3 uses cosine distance, but no theoretical or empirical justification beyond boundedness is given.
  • standard math Validation set is representative of the source distribution and large enough to learn anchors.
    Standard supervised learning assumption; the paper uses validation labels to train the anchors.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ALSA: Anchors in Logit Space for Out-of-Distribution Accuracy Estimation." pith.science (2026). https://pith.science/paper/GCH6GTNW

@misc{pith2026250819613,
  author       = {Pith},
  title        = {Pith review of: ALSA: Anchors in Logit Space for Out-of-Distribution Accuracy Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GCH6GTNW}},
  note         = {Machine review of arXiv:2508.19613}
}
read the original abstract

Estimating model accuracy on unseen, unlabeled datasets is crucial for real-world machine learning applications, especially under distribution shifts that can degrade performance. Existing methods often rely on predicted class probabilities (softmax scores) or data similarity metrics. While softmax-based approaches benefit from representing predictions on the standard simplex, compressing logits into probabilities leads to information loss. Meanwhile, similarity-based methods can be computationally expensive and domain-specific, limiting their broader applicability. In this paper, we introduce ALSA (Anchors in Logit Space for Accuracy estimation), a novel framework that preserves richer information by operating directly in the logit space. Building on theoretical insights and empirical observations, we demonstrate that the aggregation and distribution of logits exhibit a strong correlation with the predictive performance of the model. To exploit this property, ALSA employs an anchor-based modeling strategy: multiple learnable anchors are initialized in logit space, each assigned an influence function that captures subtle variations in the logits. This allows ALSA to provide robust and accurate performance estimates across a wide range of distribution shifts. Extensive experiments on vision, language, and graph benchmarks demonstrate ALSA's superiority over both softmax- and similarity-based baselines. Notably, ALSA's robustness under significant distribution shifts highlights its potential as a practical tool for reliable model evaluation.

Figures

Figures reproduced from arXiv: 2508.19613 by the authors.

Figure 1
Figure 1. Logit distributions and accuracy estimates on the Waterbirds-WILDS validation [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. The workflow of ALSA: Anchors are learned from the validation set. During the testing phase, the probability of correct￾ness is assigned to each logit. For regions where the anchor cannot generalize well, rectification is performed. Let the logit vector zi from the classifier f cor￾respond to input sample xi , that is, zi = f(xi). In order to quantify the influence that a set of anchors A exerts on the logit vector … view at source ↗
Figure 3
Figure 3. (a) MAE is reported to evaluate the GCN and SAGE models on the ogbn-arxiv [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Comparison of learnt anchors from ResNet18 on the Waterbirds-WILDS with 500 [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Visualization of the logits distribution and prediction estimations on the validation [PITH_FULL_IMAGE:figures/full_fig_p017_5.png]
Figure 6
Figure 6. Figure 6: Comparison of sampling and random initialization. [PITH_FULL_IMAGE:figures/full_fig_p018_6.png]
Figure 7
Figure 7. Figure 7: Visualization of evaluation results on CIFAR-10C and CIFAR-100C with [PITH_FULL_IMAGE:figures/full_fig_p024_7.png]
Figure 8
Figure 8. Figure 8: Visualization of the evaluation results for four GNN models: GCN, GIN, GAT, [PITH_FULL_IMAGE:figures/full_fig_p025_8.png]
Figure 9
Figure 9. Figure 9: MAE with increasing confidence interval α [PITH_FULL_IMAGE:figures/full_fig_p025_9.png]
Figure 10
Figure 10. Figure 10: MAE with increasing the number of anchors. [PITH_FULL_IMAGE:figures/full_fig_p026_10.png]
Figure 11
Figure 11. Figure 11: Inference time and MAE across different methods. [PITH_FULL_IMAGE:figures/full_fig_p026_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 48 canonical work pages

  1. [1]

    L.C. Andrews. Special Functions of Mathematics for Engineers . Online access with subscription: SPIE Digital Library. SPIE Optical Engineering Press, 1998. ISBN 9780819426161. URL https://books.google.com.au/books?id= 2CAqsF-RebgC

    work page 1998

  2. [2]

    Zico Kolter

    Christina Baek, Yiding Jiang, Aditi Raghunathan, and J. Zico Kolter. Agreement-on- the-line: Predicting the performance of neural networks under distribution shift. In NeurIPS, 2022

    work page 2022

  3. [3]

    Chen, Karan Goel, Nimit Sharad Sohoni, Fait Poms, Kayvon Fatahalian, and Christopher Ré

    Mayee F. Chen, Karan Goel, Nimit Sharad Sohoni, Fait Poms, Kayvon Fatahalian, and Christopher Ré. Mandoline: Model evaluation under distribution shift. In ICML, 2021

    work page 2021

  4. [4]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In CVPR, 2009

    2009

  5. [5]

    Are labels always necessary for classifier accuracy evaluation? In CVPR, 2021

    Weijian Deng and Liang Zheng. Are labels always necessary for classifier accuracy evaluation? In CVPR, 2021

    work page 2021

  6. [6]

    What does rotation prediction tell us about classifier accuracy under varying testing environments? In ICML, 2021

    Weijian Deng, Stephen Gould, and Liang Zheng. What does rotation prediction tell us about classifier accuracy under varying testing environments? In ICML, 2021

    work page 2021

  7. [7]

    Confidence and disper- sity speak: Characterizing prediction matrix for unsupervised accuracy estimation

    Weijian Deng, Yumin Suh, Stephen Gould, and Liang Zheng. Confidence and disper- sity speak: Characterizing prediction matrix for unsupervised accuracy estimation. In ICML, 2023

    work page 2023

  8. [8]

    Implicit generation and modeling with energy based models

    Yilun Du and Igor Mordatch. Implicit generation and modeling with energy based models. NeurIPS, 2019

    work page 2019

Show all 53 references
  1. [9]

    On out- of-distribution detection with energy-based models

    Sven Elflein, Bertrand Charpentier, Daniel Zügner, and Stephan Günnemann. On out- of-distribution detection with energy-based models. CoRR, abs/2107.08785, 2021

    2021 arXiv

  2. [10]

    Leveraging unlabeled data to predict out-of-distribution perfor- mance

    Saurabh Garg, Sivaraman Balakrishnan, Zachary Chase Lipton, Behnam Neyshabur, and Hanie Sedghi. Leveraging unlabeled data to predict out-of-distribution perfor- mance. In ICLR, 2022

    2022

  3. [11]

    Selective classification for deep neural networks

    Yonatan Geifman and Ran El-Yaniv. Selective classification for deep neural networks. In NIPS, 2017

    2017

  4. [12]

    Wich- mann, and Wieland Brendel

    Robert Geirhos, Patricia Rubisch, Claudio Michaelis, Matthias Bethge, Felix A. Wich- mann, and Wieland Brendel. Imagenet-trained cnns are biased towards texture; increas- ing shape bias improves accuracy and robustness. In ICLR, 2019

    2019

  5. [13]

    Understanding the difficulty of training deep feed- forward neural networks

    Xavier Glorot and Yoshua Bengio. Understanding the difficulty of training deep feed- forward neural networks. In AISTATS, 2010

    2010

  6. [14]

    Goodfellow, Yoshua Bengio, and Aaron C

    Ian J. Goodfellow, Yoshua Bengio, and Aaron C. Courville.Deep learning. MIT Press, 2016

    2016

  7. [15]

    Your classifier is secretly an energy based model and you should treat it like one

    Will Grathwohl, Kuan-Chieh Wang, Jörn-Henrik Jacobsen, David Duvenaud, Moham- mad Norouzi, and Kevin Swersky. Your classifier is secretly an energy based model and you should treat it like one. In ICLR, 2020. 12 LIU ET AL.: ALSA

    2020

  8. [16]

    Predicting with confidence on unseen distributions

    Devin Guillory, Vaishaal Shankar, Sayna Ebrahimi, Trevor Darrell, and Ludwig Schmidt. Predicting with confidence on unseen distributions. In ICCV, pages 1134– 1144, 2021

    2021

  9. [17]

    Weinberger

    Chuan Guo, Geoff Pleiss, Yu Sun, and Kilian Q. Weinberger. On calibration of modern neural networks. In ICML, 2017

    2017

  10. [18]

    Hamilton, Zhitao Ying, and Jure Leskovec

    William L. Hamilton, Zhitao Ying, and Jure Leskovec. Inductive representation learn- ing on large graphs. In NeurIPS, 2017

    2017

  11. [19]

    Haibo He and Edwardo A. Garcia. Learning from imbalanced data. TKDE, 2009

    2009

  12. [20]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR, 2016

    2016

  13. [21]

    Dietterich

    Dan Hendrycks and Thomas G. Dietterich. Benchmarking neural network robustness to common corruptions and perturbations. CoRR, 2019

    2019

  14. [22]

    A baseline for detecting misclassified and out-of- distribution examples in neural networks

    Dan Hendrycks and Kevin Gimpel. A baseline for detecting misclassified and out-of- distribution examples in neural networks. In ICLR, 2016

    2016

  15. [23]

    Open graph benchmark: Datasets for machine learning on graphs

    Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. Open graph benchmark: Datasets for machine learning on graphs. In NeurIPS, 2020

    2020

  16. [24]

    Guan, and Maya R

    Heinrich Jiang, Been Kim, Melody Y . Guan, and Maya R. Gupta. To trust or not to trust a classifier. In NeurIPS, 2018

    2018

  17. [25]

    Zico Kolter

    Yiding Jiang, Vaishnavh Nagarajan, Christina Baek, and J. Zico Kolter. Assessing generalization of sgd via disagreement. In ICLR, 2022

    2022

  18. [26]

    Kipf and Max Welling

    Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolu- tional networks. In ICLR, 2017

    2017

  19. [27]

    Haque, Sara M

    Pang Wei Koh, Shiori Sagawa, Henrik Marklund, Sang Michael Xie, Marvin Zhang, Akshay Balsubramani, Weihua Hu, Michihiro Yasunaga, Richard Lanas Phillips, Irena Gao, Tony Lee, Etienne David, Ian Stavness, Wei Guo, Berton Earnshaw, Imran S. Haque, Sara M. Beery, Jure Leskovec, A...

    2021

  20. [28]

    Haque, Sara M

    Pang Wei Koh, Shiori Sagawa, Henrik Marklund, Sang Michael Xie, Marvin Zhang, Akshay Balsubramani, Weihua Hu, Michihiro Yasunaga, Richard Lanas Phillips, Irena Gao, Tony Lee, Etienne David, Ian Stavness, Wei Guo, Berton Earnshaw, Imran S. Haque, Sara M. Beery, Jure Leskovec, A...

    2021

  21. [29]

    Domain adaptation by mixture of align- ments of second-or higher-order scatter tensors

    Piotr Koniusz, Yusuf Tas, and Fatih Porikli. Domain adaptation by mixture of align- ments of second-or higher-order scatter tensors. In CVPR, 2017. LIU ET AL.: ALSA 13

    2017

  22. [30]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky. Learning multiple layers of features from tiny images. Technical report, University of Toronto, 2009. URL https://www.cs.toronto.edu/ ~kriz/learning-features-2009-TR.pdf

    2009

  23. [31]

    Shiyu Liang, Yixuan Li, and R. Srikant. Enhancing the reliability of out-of-distribution image detection in neural networks. In ICLR, 2018

    2018

  24. [32]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. In ICLR, 2019

    2019

  25. [33]

    Harder or different? a closer look at distribution shift in dataset reproduction

    Shangyun Lu, Bradley Nott, Aaron Olson, Alberto Todeschini, Hossein Vahabi, Yair Carmon, and Ludwig Schmidt. Harder or different? a closer look at distribution shift in dataset reproduction. In ICML Workshop on Uncertainty and Robustness in Deep Learning, volume 5, page 15, 2020

    2020

  26. [34]

    Yuzhe Lu, Yilong Qin, Runtian Zhai, Andrew Shen, Ketong Chen, Zhenlin Wang, So- heil Kolouri, Simon Stepputtis, Joseph Campbell, and Katia P. Sycara. Characterizing out-of-distribution error via optimal transport. NeurIPS, 2023

    2023

  27. [35]

    Accuracy on the line: on the strong correlation between out-of-distribution and in-distribution generalization

    John Miller, Rohan Taori, Aditi Raghunathan, Shiori Sagawa, Pang Wei Koh, Vaishaal Shankar, Percy Liang, Yair Carmon, and Ludwig Schmidt. Accuracy on the line: on the strong correlation between out-of-distribution and in-distribution generalization. In ICML, 2021

    2021

  28. [36]

    Distributional generalization: A new kind of generalization

    Preetum Nakkiran and Yamini Bansal. Distributional generalization: A new kind of generalization. CoRR, abs/2009.08092, 2020

    2009 arXiv

  29. [37]

    Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chin- tala

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Des- maison, Andreas Köpf, Edward Z. Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner...

    2019

  30. [38]

    Torchvision models

    PyTorch. Torchvision models. https://pytorch.org/vision/main/ models.html, 2024. Accessed: 2024-09-21

    2024

  31. [39]

    Do cifar- 10 classifiers generalize to cifar-10? CoRR, 2018

    Benjamin Recht, Rebecca Roelofs, Ludwig Schmidt, and Vaishaal Shankar. Do cifar- 10 classifiers generalize to cifar-10? CoRR, 2018

    2018

  32. [40]

    Do ima- genet classifiers generalize to imagenet? In ICML, 2019

    Benjamin Recht, Rebecca Roelofs, Ludwig Schmidt, and Vaishaal Shankar. Do ima- genet classifiers generalize to imagenet? In ICML, 2019

    2019

  33. [41]

    Hashimoto, and Percy Liang

    Shiori Sagawa, Pang Wei Koh, Tatsunori B. Hashimoto, and Percy Liang. Distribution- ally robust neural networks for group shifts: On the importance of regularization for worst-case generalization. CoRR, 2019

    2019

  34. [42]

    Distilbert, a dis- tilled version of bert: Smaller, faster, cheaper and lighter

    Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. Distilbert, a dis- tilled version of bert: Smaller, faster, cheaper and lighter. CoRR, 2019

    2019

  35. [43]

    Long-tailed classification by keeping the good and removing the bad momentum causal effect

    Kaihua Tang, Jianqiang Huang, and Hanwang Zhang. Long-tailed classification by keeping the good and removing the bad momentum causal effect. In NeurIPS, 2020. 14 LIU ET AL.: ALSA

    2020

  36. [44]

    Graph attention networks

    Petar Velickovic, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. Graph attention networks. In ICLR, 2018

    2018

  37. [45]

    GOLD: graph out-of- distribution detection via implicit adversarial latent generation

    Danny Wang, Ruihong Qiu, Guangdong Bai, and Zi Huang. GOLD: graph out-of- distribution detection via implicit adversarial latent generation. In ICLR, 2025

    2025

  38. [46]

    Text meets topology: Rethinking out-of-distribution detection in text-rich networks

    Danny Wang, Ruihong Qiu, Guangdong Bai, and Zi Huang. Text meets topology: Rethinking out-of-distribution detection in text-rich networks. CoRR, abs/2508.17690, 2025

    2025 arXiv

  39. [47]

    How powerful are graph neural networks? In ICLR, 2018

    Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural networks? In ICLR, 2018

    2018

  40. [48]

    Predicting out-of-distribution error with the projection norm

    Yaodong Yu, Zitong Yang, Alexander Wei, Yi Ma, and Jacob Steinhardt. Predicting out-of-distribution error with the projection norm. In ICML, 2022

    2022

  41. [49]

    Deep structured energy based models for anomaly detection

    Shuangfei Zhai, Yu Cheng, Weining Lu, and Zhongfei Zhang. Deep structured energy based models for anomaly detection. In ICML, 2016

    2016

  42. [50]

    Hybrid models for open set recog- nition

    Hongjie Zhang, Ang Li, Jie Guo, and Yanwen Guo. Hybrid models for open set recog- nition. In ECCV, 2020

    2020

  43. [51]

    Gonzalez, Alberto L

    Sicheng Zhao, Xiangyu Yue, Shanghang Zhang, Bo Li, Han Zhao, Bichen Wu, Ravi Krishna, Joseph E. Gonzalez, Alberto L. Sangiovanni-Vincentelli, Sanjit A. Seshia, and Kurt Keutzer. A review of single-source deep unsupervised visual domain adaptation. IEEE Trans. Neural Networks L...

    2022

  44. [52]

    Gnnevaluator: Evaluating GNN performance on unseen graphs without labels

    Xin Zheng, Miao Zhang, Chunyang Chen, Soheila Molaei, Chuan Zhou, and Shirui Pan. Gnnevaluator: Evaluating GNN performance on unseen graphs without labels. In NeurIPS, 2023. LIU ET AL.: ALSA 15 Appendix A Related Work This section reviews methods of detecting out-of-distributi...

    2023

  45. [2020]

    I Detailed Experiment Results I.1 Full Table of Vision and Language Datasets Results Comprehensive experiments were conducted on vision and language datasets

    For all models, the Adam optimizer was used with a fixed learning rate of 10 −2 and a weight decay of 5 × 10−4, along with early stopping if validation accuracy did not improve after 20 epochs. I Detailed Experiment Results I.1 Full Table of Vision and Language Datasets Result...

    2015

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.