Out-of-Distribution Detection with Prototypical Outlier Proxy

Chaoqi Chen; Hui Huang; Mingrong Gong; Qingqiang Sun; Yue Wang

REVIEW 3 major objections 4 minor 51 references

Out-of-Distribution Detection with Prototypical Outlier Proxy

T0 review · 3 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Prototypical outlier proxies are virtual OOD anchors added to a fixed, hierarchy-aware classifier, reshaping decision boundaries so deep models stop being overconfident on unseen data without ever seeing a real or synthetic outlier.

desk verdict A simple, fast OOD detector with a promising idea, but the headline margins are at risk from test-set hyperparameter tuning and missing error bars. read the letter →

arxiv 2412.16884 v1 pith:JZXJOH4Z submitted 2024-12-22 cs.CV

Mingrong Gong , Chaoqi Chen , Qingqiang Sun , Yue Wang , Hui Huang This is my paper · ORCID

classification cs.CV

keywords OODdetectionout-of-distributionprototypicaloutlierproxyfixedclassifierhierarchicalsimilarityboundarylossneuralcollapsecosinevirtual

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

The paper claims that deep models can be made to detect out-of-distribution (OOD) data without being shown any real or synthesized outliers. Its method, POP, replaces the learnable classifier with a fixed cosine classifier whose weight vectors come from the semantic hierarchy of the training classes, then appends a few prototypical weight vectors that act as virtual OOD centers. A hierarchical similarity boundary loss (HSBL) applies stronger penalties to more severe misclassifications, and the OOD score is feature-norm-scaled max logit. On the standardized OOD benchmark, POP reports average FPR95 reductions of 7.70% on CIFAR-10, 6.30% on CIFAR-100, and 5.42% on ImageNet-200 over the second-best baselines, while training 7.2 times and inferring 19.5 times faster than the outlier-synthesis method NPOS. A sympathetic reader would care because this suggests a cheap, distribution-free route to OOD awareness.

What carries the argument

The load-bearing construction is the hierarchy-aware fixed classifier: build the hierarchy distance matrix $D$ where $d_{ij}$ is the height of the lowest common ancestor of classes $y_i$ and $y_j$, convert it to similarity via $\varphi(d_{ij}) = 1/(d_{ij}+1)$, and factor the resulting symmetric matrix $S$ through eigen-decomposition and QR decomposition to obtain unit weight vectors $W$ with $W^T W = S$. POP extends $D$ to $D_{\mathrm{pop}}$ by appending $C$ rows and columns of a constant OOD distance $d$, repeats the factorization, and concatenates the resulting outlier-proxy weights to the ID prototypes. The hierarchical similarity boundary loss replaces the plain cross-entropy logit with $\beta$-scaled cosine logits minus a margin $m_{\hat{y}y} = 1 - s_{\hat{y}y}$ for wrong predictions, so penalties scale with semantic mismatch. The OOD score $S(X) = \lVert x \rVert \cdot \max(z)$ combines the feature norm with the largest cosine logit.

What would settle it

Compute the eigenvalues of $S_{\mathrm{pop}}$ for the CIFAR-100 and ImageNet-200 hierarchies with the $C$ and $d$ values reported in the paper's configuration table. If any eigenvalue is negative to numerical precision, Eqs. (1)-(2) cannot yield a real weight matrix $W_{\mathrm{pop}}$, and the POP pipeline as described does not run; conversely, if all eigenvalues are nonnegative, the core construction is reproducible.

Watch

Extended reading notes

Core claim

POP's central claim is that the overconfidence of deep classifiers on OOD inputs can be mitigated by geometry rather than data: by fixing the final linear layer to a precomputed set of unit prototype vectors that respect the class hierarchy, and by adjoining $C$ prototypical outlier proxies at a uniform distance beyond the farthest ID classes, the feature space is carved into tighter ID regions with explicit unknown zones. Training with HSBL then enforces cosine margins proportional to semantic distance: a sample confused with a semantically close class is penalized little, while confusion with a distant class or an outlier proxy is penalized heavily. At test time, multiplying the max logit by the feature norm gives the OOD score. The paper's evidence is a set of benchmark results, ablations, and timing comparisons; on its terms, POP shows that outlier proxies with no real outlier exposure can match or beat methods that train on synthetic outliers.

Load-bearing premise

The construction assumes the extended similarity matrix $S_{\mathrm{pop}}$, built from hierarchy distances through $\varphi(d)=1/(d+1)$, is positive semidefinite, since the fixed classifier weights come from factoring $S_{\mathrm{pop}}$; if that fails for CIFAR-100 or ImageNet-200, the method cannot be implemented as written.

Editorial extensions

If this is right

If POP is right, OOD detection no longer needs outlier exposure or density estimation, removing a major training cost and avoiding bias toward particular outlier types.
The reported speedups (7.2x training, 19.5x inference vs NPOS) imply that OOD-aware training can be used in real-time and resource-constrained settings.
Adaptive semantic penalties from HSBL should transfer to any task with a class taxonomy, such as fine-grained recognition or autonomous driving, where mistake severity varies.
Combining fixed prototypes with outlier proxies yields a balance between near-OOD and far-OOD performance, addressing a failure mode of outlier-exposure methods on distant data like MNIST.
Because the score function needs no ID data at test time, deployment is simpler and avoids privacy and data-leakage concerns.

Reading between the lines

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

Editorial inference: the fixed-proxy construction should generalize to any tree-structured label space; a cheap test is to run POP on full ImageNet-1k or on domain-specific taxonomies and check whether the FPR95 gains persist beyond the 200-class subset studied here.
Editorial inference: the paper's weaker result on CIFAR-10 as ID with CIFAR-100 as near-OOD suggests the gains concentrate when ID itself is fine-grained; one could test whether HSBL margins should be normalized by tree depth to avoid over-penalizing coarse datasets.
Editorial inference: because the outlier proxies sit at a single constant distance, their positions are isotropic; weighting their distances by local ID feature density, or making them class-conditional, may further reduce false positives without introducing real outliers.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Good paper to know about. The idea is straightforward: take a hierarchy-fixed classifier, append a small set of fixed 'outlier prototype' weight vectors at a chosen distance, and train with a margin loss that penalizes misclassifications by hierarchy distance. It removes the need for outlier exposure or synthetic outlier generation, and the efficiency gains over NPOS are real and meaningful. On the evidence in the paper, the method is consistent across three benchmarks and the ablations support each component's contribution. This is a genuinely practical contribution, and anyone looking for a cheap, effective OOD detector without outlier data will want this as a baseline.

The soft spots are the usual ones for training-time OOD papers. The hyperparameters d, C, and beta are selected by grid search on the OOD test sets themselves (Fig. 6c, Tab. 9). No separate validation split is described. That can inflate the reported margins, especially the 5-8 point FPR95 drops over second-best. The authors should rerun with a proper validation split and report error bars. This is fixable, and I would not call the central idea wrong, but the exact magnitudes in the abstract should be treated with caution.

The PSD issue in Eqs. 1-5 is a rigor gap, but I agree with the stress-tester that it is not the main risk. For these ultrametrics and the completely monotone mapping, Spop is likely PSD via an integral representation; a proof would settle it, but it is not load-bearing.

The writing is clear, the related work is fair, and the code is promised. This deserves a serious referee, but the revision needs a validation protocol and uncertainty estimates before the numbers can be trusted as stated.

Referee Report

3 major / 4 minor

Summary. The paper proposes Prototypical Outlier Proxy (POP), a training-time OOD detection method that replaces the learnable classifier with a fixed classifier derived from a semantic hierarchy (HAFrame), appends C fixed virtual outlier-proxy weight vectors at a chosen distance d from all ID prototypes, and trains the feature extractor with a hierarchical similarity boundary loss (HSBL) that applies larger penalties to more semantically distant misclassifications. At test time, the OOD score is the feature-norm-scaled max logit. Experiments on CIFAR-10, CIFAR-100, and ImageNet-200 under the OpenOOD protocol report average FPR95 reductions of 7.70%, 6.30%, and 5.42% over the second-best baselines, with substantial speedups over the synthetic-outlier method NPOS. Ablations attribute the gains to the fixed hierarchical classifier, the outlier proxies, and HSBL.

Significance. If the empirical claims survive scrutiny, POP is an appealing training-time alternative to outlier exposure and outlier synthesis: it requires no real or generated OOD samples, is simple to implement, and is substantially faster than NPOS. The conceptual move of building virtual outlier proxies into a fixed hierarchical classifier is original and the code release will make the method easy to reproduce. The main caveats are that the headline margins are obtained under a hyperparameter-selection protocol that does not use a described validation split, no error bars are reported for the central tables, and the mathematical construction hinges on an unproven positive-semidefiniteness claim. These issues are addressable but currently leave the strength of the contribution uncertain.

major comments (3)

[Impact of Prototypical Outlier Proxies and Appendix B.2] The headline improvement is not protected against test-set overfitting. In the section on the impact of prototypical outlier proxies, the grid search over d and C is evaluated on the OOD test benchmarks (Fig. 6(c)), and Appendix B.2 (Tables 8 and 9) selects the final values of d, C, and beta using the same test benchmarks, with no separate validation split described anywhere in the paper. Because the central claim is the average FPR95 reduction over the second-best methods, the reported margins may be optimistic. Please repeat the main comparisons with hyperparameters chosen on a validation split (or with a nested protocol) and report both validation-selected and test-set results, or at least provide a sensitivity analysis showing that the gains persist across the grid.
[Eqs. (1)-(5)] The construction of W_pop relies on the eigendecomposition S = Q P Q^T and on forming U P^{1/2} Q^T, which requires the extended similarity matrix S_pop to be positive semidefinite. The paper does not prove that the hierarchy-based distance matrix augmented with a constant-distance outlier block, mapped through phi(d) = 1/(d+1), is PSD for the CIFAR-100 and ImageNet-200 hierarchies or for arbitrary C. This is load-bearing: if S_pop has negative eigenvalues, the real-valued classifier weights W_pop do not exist and the method cannot be implemented as described. Please add a proof or a rigorous argument (for example, using complete monotonicity of phi and ultrametricity of the distance matrix) and state explicitly the conditions under which the construction is valid.
[Tables 1-3] The main OOD detection results are reported as single point estimates with no standard deviations, confidence intervals, or number of seeds. Since several claims are margins of roughly 5 to 8 percentage points in FPR95 over the second-best baseline, and since the hyperparameters are selected on the test benchmarks, the reader cannot assess whether the reported differences are statistically reliable. Please report mean and standard deviation over at least three independent training runs for the main tables, and, where feasible, paired comparisons against the strongest baselines.

minor comments (4)

[Eq. (7)] The angle notation in Eq. (7) is inconsistent: in Eq. (6), theta_{i,j} is defined as the angle between weight vectors w_i and w_j, but in Eq. (7) the denominator should contain the angle between the feature vector x_i and the weight vector w_j. Please clarify by writing cos(theta_{j,i}) or by redefining the notation explicitly.
[Experiments: Results on CIFAR-10 and CIFAR-100] Several numerical statements in the text do not match the tables. For CIFAR-10, the text reports a 5.4% FPR95 reduction and a 2.05% AUROC improvement, but Table 1 gives a 7.70% FPR95 reduction and a 2.65% AUROC improvement over the second-best baseline (NPOS). For CIFAR-100, the text reports a 3.52% AUROC improvement, but Table 2 gives a 2.52% improvement over the best baseline (KNN). Please reconcile the prose with the tables.
[Table 1 caption] The Table 1 caption mentions 'ID Accuracy' as a reported metric, but the table columns contain only FPR95 and AUROC. Please either add the ID accuracy column or remove that mention.
[Appendix B] Appendix B has two subsections labeled B.2: 'B.2 Training Details' and 'B.2 Effect of beta'. Please renumber the second subsection to avoid confusion.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the classifier and loss are constructed from the ID hierarchy, and the reported gains are external-benchmark results, though test-set hyperparameter tuning is a soundness caveat.

full rationale

The derivation chain is self-contained: Wpop is obtained by applying a fixed spectral/QR decomposition (Eqs. 1–5) to the similarity matrix Spop = phi(Dpop), where Dpop is built from the ID label hierarchy plus a chosen constant distance d; the HSBL margins m = 1 - s are the same hierarchical similarities, and the test-time score is MaxLogit scaled by feature norm. No equation in this chain uses OOD test data or a parameter fitted to the evaluation metric, so no step reduces by construction to its own input. The main caveats are evaluation-protocol concerns rather than circularity: d, C, and beta are selected by grid search against the OOD test benchmarks (Fig. 6c, Tab. 8, Tab. 9), which means the reported FPR95 reductions may be optimistically biased and should be confirmed on held-out validation splits; and the PSD of Spop for CIFAR-100/ImageNet-200 is asserted via eigen-decomposition without a proof. One peripheral self-citation (Chen et al. 2023a in the HSBL analysis) is illustrative, not load-bearing. Accordingly, no circular derivation is exhibited.

Assumptions & free parameters 3 free parameters · 3 assumptions · 1 invented entities

The central method depends on three hand-set hyperparameters, a positive-semidefiniteness assumption for the similarity kernel, and the assumption that a semantic label tree is a reliable proxy for feature-space geometry. The only invented entity is the prototypical outlier proxy, which is a model component rather than a physical observation, and it lacks independent external evidence beyond the reported benchmark numbers.

free parameters (3)

OOD distance d = CIFAR-10: 4, CIFAR-100: 7, ImageNet-200: 18
Controls the placement of outlier proxies relative to ID prototypes in the distance matrix Dpop (Eq. 3). Chosen by grid search on CIFAR-10 and by unspecified tuning on the other datasets; directly affects the decision boundary and the final performance.
Number of outlier proxies C = CIFAR-10: 2, CIFAR-100: 60, ImageNet-200: 40
Number of added virtual OOD weight vectors; selected by grid search or tuning. Too few proxies may under-cover the feature space, while too many hurt performance (Fig. 6c).
HSBL scaling factor beta = CIFAR-10: 10, CIFAR-100: 5, ImageNet-200: 10
Temperature in the hierarchical similarity boundary loss (Eq. 7); tuned per dataset as shown in Table 9, with no separate validation split described.

assumptions (3)

domain assumption The semantic hierarchy distance matrix D and its extension Dpop yield a positive semidefinite similarity matrix S = phi(D) with phi(d)=1/(d+1), so that the fixed classifier weights Wpop can be obtained via eigen-decomposition (Eqs. 1-2).
The existence of the fixed classifier relies entirely on this decomposition. The paper does not prove PSD for the chosen kernel on the CIFAR-100 and ImageNet-200 trees, nor for the appended constant-distance outlier block. If S is not PSD, the construction fails.
domain assumption The label hierarchy (WordNet-style tree) is provided for each dataset and reflects the feature-space geometry of the in-distribution classes.
The fixed prototypes are positioned according to the hierarchy; if the tree does not meaningfully align with the feature manifold, the method's decision boundaries will be misaligned with the data distribution.
domain assumption Training with HSBL drives the feature extractor to align with the fixed prototype geometry (a neural-collapse-like behavior).
The test-time score assumes that ID features cluster near their fixed prototypes and that OOD features do not. This is an empirical outcome of training and is not guaranteed by the loss construction, though the paper's toy experiment and ablations suggest it happens in practice.

invented entities (1)

Prototypical outlier proxy weight vectors
purpose: Added to the frozen classifier as virtual OOD class centers to attract OOD features and compress ID decision boundaries without requiring real or synthetic outlier examples.
These vectors are internal model parameters, not observed phenomena. Their utility is inferred indirectly from benchmark FPR95 and AUROC gains, and they have no falsifiable handle outside the paper's own experiments.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Out-of-Distribution Detection with Prototypical Outlier Proxy." pith.science (2026). https://pith.science/paper/JZXJOH4Z

@misc{pith2026241216884,
  author       = {Pith},
  title        = {Pith review of: Out-of-Distribution Detection with Prototypical Outlier Proxy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JZXJOH4Z}},
  note         = {Machine review of arXiv:2412.16884}
}

read the original abstract

Out-of-distribution (OOD) detection is a crucial task for deploying deep learning models in the wild. One of the major challenges is that well-trained deep models tend to perform over-confidence on unseen test data. Recent research attempts to leverage real or synthetic outliers to mitigate the issue, which may significantly increase computational costs and be biased toward specific outlier characteristics. In this paper, we propose a simple yet effective framework, Prototypical Outlier Proxy (POP), which introduces virtual OOD prototypes to reshape the decision boundaries between ID and OOD data. Specifically, we transform the learnable classifier into a fixed one and augment it with a set of prototypical weight vectors. Then, we introduce a hierarchical similarity boundary loss to impose adaptive penalties depending on the degree of misclassification. Extensive experiments across various benchmarks demonstrate the effectiveness of POP. Notably, POP achieves average FPR95 reductions of 7.70%, 6.30%, and 5.42% over the second-best methods on CIFAR-10, CIFAR-100, and ImageNet-200, respectively. Moreover, compared to the recent method NPOS, which relies on outlier synthesis, POP trains 7.2X faster and performs inference 19.5X faster. The source code is available at: https://github.com/gmr523/pop.

Figures

Figures reproduced from arXiv: 2412.16884 by the authors.

**Figure 1.** Illustration of our motivation. Left: Vanilla training. Middle: Training with the mixture of ID data and outliers. Right: Training with prototypical outlier proxies. exposure (OE) (Hendrycks, Mazeika, and Dietterich 2019; Yu and Aizawa 2019; Yang et al. 2021; Ming, Fan, and Li 2022; Zhang et al. 2023b), or feature-based outlier synthesis (Pei et al. 2022; Du et al. 2022; Tao et al. 2023) [PITH_FULL_IMAGE:figures/f… view at source ↗

**Figure 3.** Toy example. Use a ResNet-18 with a feature layer size of 2 for three CIFAR-10 classes. The x- and yaxes represent the feature values in the square region. We evaluate prediction confidence for each point in these classes. Yellow, green, and blue points represent deer, horse, and ship, respectively. (a) Vanilla ResNet-18. (b) Fixed ResNet-18. (c) Fixed ResNet-18 with one outlier proxy (brown star marker). Motivatio… view at source ↗

**Figure 4.** The overview of POP. The contributions module [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗

Figures from the paper (3 more)

**Figure 5.** Figure 5: The principles of HSBL x such that ||x|| = 1, where x ∈ RM denotes the feature vector of the ID data. For the i th ID data’s feature xi with ground-truth label yi and predicted label yˆi , the cross-entropy (CE) loss Lce in cosine space is: Lce = − N X +C i log e wT yi…

**Figure 6.** Figure 6: The analysis of HSBL loss and POP. ID test data (CIFAR-10) and OOD data (SVHN) features are visualized using [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

**Figure 7.** Figure 7: The hierarchical structure of CIFAR-10. The maximum hierarchical distances dmax for CIFAR-10, CIFAR-100, and ImageNet-200 are presented in Tab. 7. POP calculates the OOD distance based on dmax, thereby introducing outlier prototypes. CIFAR-10 CIFAR-100 ImageNet-200 dma…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 40 canonical work pages

[1]

, " * write output.state after.block = add.period write newline

ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...
[2]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...
[3]

Amodei, D.; Olah, C.; Steinhardt, J.; Christiano, P.; Schulman, J.; and Man \'e , D. 2016. Concrete problems in AI safety. arXiv preprint arXiv:1606.06565

arXiv 2016
[4]

Bitterwolf, J.; M \" u ller, M.; and Hein, M. 2023. In or Out? Fixing ImageNet Out-of-Distribution Detection Evaluation. In ICML , volume 202, 2471--2506

work page 2023
[5]

Chen, C.; Tang, L.; Huang, Y.; Han, X.; and Yu, Y. 2023 a . CODA: Generalizing to Open and Unseen Domains with Compaction and Disambiguation. In NeurIPS, volume 36, 12746--12759

work page 2023
[6]

Chen, L.; Wu, P.; Chitta, K.; Jaeger, B.; Geiger, A.; and Li, H. 2023 b . End-to-end autonomous driving: Challenges and frontiers. arXiv preprint arXiv:2306.16927

arXiv 2023
[7]

Cimpoi, M.; Maji, S.; Kokkinos, I.; Mohamed, S.; and Vedaldi, A. 2014. Describing textures in the wild. In CVPR, 3606--3613

work page 2014
[8]

Deng, J.; Dong, W.; Socher, R.; Li, L.; Li, K.; and Fei - Fei, L. 2009. ImageNet: A large-scale hierarchical image database. In CVPR , 248--255

work page 2009

Show all 51 references

[9]

Deng, J.; Guo, J.; Xue, N.; and Zafeiriou, S. 2019. Arcface: Additive angular margin loss for deep face recognition. In CVPR, 4690--4699

2019
[10]

Deng, L. 2012. The MNIST Database of Handwritten Digit Images for Machine Learning Research [Best of the Web]. IEEE Signal Processing Magazine, 29(6): 141--142

2012
[11]

Du, X.; Sun, Y.; Zhu, J.; and Li, Y. 2023. Dream the Impossible: Outlier Imagination with Diffusion Models. In NeurIPS, volume 36, 60878--60901

2023
[12]

Du, X.; Wang, Z.; Cai, M.; and Li, Y. 2022. VOS: Learning What You Don't Know by Virtual Outlier Synthesis. In ICLR

2022
[13]

Fang, Z.; Li, Y.; Lu, J.; Dong, J.; Han, B.; and Liu, F. 2022. Is out-of-distribution detection learnable? In NeurIPS, volume 35, 37199--37213

2022
[14]

Fellbaum, C. 1998. WordNet: An electronic lexical database. MIT press

1998
[15]

Garg, A.; Sani, D.; and Anand, S. 2022. Learning Hierarchy Aware Features for Reducing Mistake Severity. In ECCV , volume 13684, 252--267

2022
[16]

Guo, C.; Pleiss, G.; Sun, Y.; and Weinberger, K. Q. 2017. On Calibration of Modern Neural Networks. In ICML , volume 70, 1321--1330

2017
[17]

Hardt, M.; and Ma, T. 2017. Identity Matters in Deep Learning. In ICLR

2017
[18]

He, K.; Gkioxari, G.; Doll \' a r, P.; and Girshick, R. B. 2017. Mask R-CNN . In ICCV , 2980--2988

2017
[19]

He, K.; Zhang, X.; Ren, S.; and Sun, J. 2016. Deep Residual Learning for Image Recognition. In CVPR , 770--778

2016
[20]

Hendrycks, D.; Basart, S.; Mazeika, M.; Mostajabi, M.; Steinhardt, J.; and Song, D. X. 2022. Scaling Out-of-Distribution Detection for Real-World Settings. In ICML, volume 162, 8759--8773

2022
[21]

Hendrycks, D.; and Gimpel, K. 2017. A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks. In ICLR

2017
[22]

Hendrycks, D.; Mazeika, M.; and Dietterich, T. G. 2019. Deep Anomaly Detection with Outlier Exposure. In ICLR

2019
[23]

Hoffer, E.; Hubara, I.; and Soudry, D. 2018. Fix your classifier: the marginal value of training the last weight layer. In ICLR

2018
[24]

Huang, S.; and Mo, J. 2024. Fixed Random Classifier Rearrangement for Continual Learning. arXiv:2402.15227

2024 arXiv
[25]

Krizhevsky, A.; Hinton, G.; et al. 2009. Learning multiple layers of features from tiny images

2009
[26]

Lee, K.; Lee, K.; Lee, H.; and Shin, J. 2018. A simple unified framework for detecting out-of-distribution samples and adversarial attacks. In NeurIPS, volume 31, 7167--7177

2018
[27]

Levine, S.; Finn, C.; Darrell, T.; and Abbeel, P. 2016. End-to-end training of deep visuomotor policies. Journal of Machine Learning Research, 17(1): 1334--1373

2016
[28]

Liang, T.; and Davis, J. 2023. Inducing Neural Collapse to a Fixed Hierarchy-Aware Frame for Reducing Mistake Severity. In ICCV , 1443--1452

2023
[29]

D.; and Li, Y

Liu, W.; Wang, X.; Owens, J. D.; and Li, Y. 2020. Energy-based Out-of-distribution Detection. In NeurIPS, volume 33, 21464--21475

2020
[30]

Ma, Y.; Tsao, D.; and Shum, H.-Y. 2022. On the principles of parsimony and self-consistency for the emergence of intelligence. Frontiers of Information Technology & Electronic Engineering, 23(9): 1298--1323

2022
[31]

McInnes, L.; Healy, J.; and Melville, J. 2020. UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction. arXiv:1802.03426

2020 arXiv
[32]

Ming, Y.; Fan, Y.; and Li, Y. 2022. Poem: Out-of-distribution detection with posterior sampling. In ICML, volume 162, 15650--15665

2022
[33]

Ming, Y.; Yin, H.; and Li, Y. 2022. On the impact of spurious correlation for out-of-distribution detection. In Proceedings of the AAAI conference on artificial intelligence, volume 36, 10051--10059

2022
[34]

Nie, J.; Luo, Y.; Ye, S.; Zhang, Y.; Tian, X.; and Fang, Z. 2024. Out-of-distribution detection with virtual outlier smoothing. International Journal of Computer Vision, 1--18

2024
[35]

Papyan, V.; Han, X.; and Donoho, D. L. 2020. Prevalence of neural collapse during the terminal phase of deep learning training. Proceedings of the National Academy of Sciences, 117(40): 24652--24663

2020
[36]

Pei, S.; Zhang, X.; Fan, B.; and Meng, G. 2022. Out-of-distribution Detection with Boundary Aware Learning. In ECCV , volume 13684, 235--251

2022
[37]

Sun, Y.; Ming, Y.; Zhu, X.; and Li, Y. 2022. Out-of-Distribution Detection with Deep Nearest Neighbors. In ICML , volume 162, 20827--20840

2022
[38]

Tao, L.; Du, X.; Zhu, J.; and Li, Y. 2023. Non-parametric Outlier Synthesis. In ICLR

2023
[39]

Torralba, A.; Fergus, R.; and Freeman, W. T. 2008. 80 million tiny images: A large data set for nonparametric object and scene recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 30(11): 1958--1970

2008
[40]

Van Horn, G.; Mac Aodha, O.; Song, Y.; Cui, Y.; Sun, C.; Shepard, A.; Adam, H.; Perona, P.; and Belongie, S. 2018. The inaturalist species classification and detection dataset. In CVPR, 8769--8778

2018
[41]

Vaze, S.; Han, K.; Vedaldi, A.; and Zisserman, A. 2022. Open-Set Recognition: A Good Closed-Set Classifier is All You Need. In ICLR

2022
[42]

Wang, H.; Li, Z.; Feng, L.; and Zhang, W. 2022. Vim: Out-of-distribution with virtual-logit matching. In CVPR, 4921--4930

2022
[43]

Xiao, H.; Rasul, K.; and Vollgraf, R. 2017. Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms. arXiv preprint arXiv:1708.07747

2017 arXiv
[44]

Yang, J.; Wang, H.; Feng, L.; Yan, X.; Zheng, H.; Zhang, W.; and Liu, Z. 2021. Semantically Coherent Out-of-Distribution Detection. In ICCV , 8281--8289

2021
[45]

Yang, J.; Wang, P.; Zou, D.; Zhou, Z.; Ding, K.; Peng, W.; Wang, H.; Chen, G.; Li, B.; Sun, Y.; et al. 2022 a . Openood: Benchmarking generalized out-of-distribution detection. In NeurIPS, volume 35, 32598--32611

2022
[46]

Yang, Y.; Chen, S.; Li, X.; Xie, L.; Lin, Z.; and Tao, D. 2022 b . Inducing Neural Collapse in Imbalanced Learning: Do We Really Need a Learnable Classifier at the End of Deep Neural Network? In NeurIPS, volume 35, 37991--38002

2022
[47]

Yu, Q.; and Aizawa, K. 2019. Unsupervised Out-of-Distribution Detection by Maximum Classifier Discrepancy. In ICCV, 9517--9525

2019
[48]

Zhang, J.; Gao, L.; Hao, B.; Huang, H.; Song, J.; and Shen, H. 2023 a . From Global to Local: Multi-Scale Out-of-Distribution Detection. IEEE Trans. Image Process. , 32: 6115--6128

2023
[49]

Zhang, J.; Inkawhich, N.; Linderman, R.; Chen, Y.; and Li, H. 2023 b . Mixture Outlier Exposure: Towards Out-of-Distribution Detection in Fine-grained Environments. In WACV , 5520--5529

2023
[50]

Zhang, J.; Yang, J.; Wang, P.; Wang, H.; Lin, Y.; Zhang, H.; Sun, Y.; Du, X.; Zhou, K.; Zhang, W.; Li, Y.; Liu, Z.; Chen, Y.; and Li, H. 2023 c . OpenOOD v1.5: Enhanced Benchmark for Out-of-Distribution Detection. arXiv preprint arXiv:2306.09301

2023 arXiv
[51]

Zhou, B.; Khosla, A.; Lapedriza, A.; Torralba, A.; and Oliva, A. 2016. Places: An image database for deep scene understanding. arXiv preprint arXiv:1610.02055

2016 arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

, " * write output.state after.block = add.period write newline

ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

[2] [2]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

[3] [3]

Amodei, D.; Olah, C.; Steinhardt, J.; Christiano, P.; Schulman, J.; and Man \'e , D. 2016. Concrete problems in AI safety. arXiv preprint arXiv:1606.06565

arXiv 2016

[4] [4]

Bitterwolf, J.; M \" u ller, M.; and Hein, M. 2023. In or Out? Fixing ImageNet Out-of-Distribution Detection Evaluation. In ICML , volume 202, 2471--2506

work page 2023

[5] [5]

Chen, C.; Tang, L.; Huang, Y.; Han, X.; and Yu, Y. 2023 a . CODA: Generalizing to Open and Unseen Domains with Compaction and Disambiguation. In NeurIPS, volume 36, 12746--12759

work page 2023

[6] [6]

Chen, L.; Wu, P.; Chitta, K.; Jaeger, B.; Geiger, A.; and Li, H. 2023 b . End-to-end autonomous driving: Challenges and frontiers. arXiv preprint arXiv:2306.16927

arXiv 2023

[7] [7]

Cimpoi, M.; Maji, S.; Kokkinos, I.; Mohamed, S.; and Vedaldi, A. 2014. Describing textures in the wild. In CVPR, 3606--3613

work page 2014

[8] [8]

Deng, J.; Dong, W.; Socher, R.; Li, L.; Li, K.; and Fei - Fei, L. 2009. ImageNet: A large-scale hierarchical image database. In CVPR , 248--255

work page 2009