Pith. sign in

REVIEW 2 major objections 3 minor 45 references

Detecting semantic anomalies

T0 review · 2 major / 3 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Practical out-of-distribution detection is semantic anomaly detection, and self-supervised auxiliary tasks improve it.

desk verdict Useful corrective to sloppy OOD benchmarks; the rotation/CPC gains are real on CIFAR/STL-10 but the ImageNet subset evidence is too thin to support the abstract's broad claim. read the letter →

arxiv 1908.04388 v3 pith:2OXYEMWA submitted 2019-08-13 cs.CV cs.LG

Faruk Ahmed , Aaron Courville This is my paper · ORCID
classification cs.CVcs.LG
keywords out-of-distributiondetectionsemanticanomalyhold-out-classevaluationmulti-tasklearningself-supervisedrotationpredictioncontrastivepredictivecodingobjectrecognition
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

Most out-of-distribution detection benchmarks treat whole datasets as out-distributions, where low-level statistics such as color or edge density are enough to separate them from the training set. This paper contends that practically relevant novelty is semantic: in a specified context such as object recognition, a shift in object identity should be flagged, while non-semantic variation such as lighting or compression should not. To evaluate that claim, it recommends holding out one class at a time on CIFAR-10, STL-10, and five fine-grained ImageNet subsets, and shows that training classifiers with auxiliary self-supervised objectives, predicting image rotation or contrastive predictive coding, improves anomaly-detection average precision while also improving classification accuracy. The authors conclude that semantic anomaly detection is the well-motivated form of out-of-distribution detection, and that multi-task learning with such objectives is a promising way to induce it.

What carries the argument

The central mechanism is multi-task learning with an auxiliary self-supervised objective. The combined training loss is $\mathcal{L}_{\text{combined}}(\theta;D) = \mathcal{L}_{\text{primary}}(\theta;D) + \lambda \mathcal{L}_{\text{auxiliary}}(\theta;D)$, where $\mathcal{L}_{\text{primary}}$ is the categorical cross-entropy of object classification and $\mathcal{L}_{\text{auxiliary}}$ is either rotation prediction, predicting which of a fixed set of rotations was applied to an image, or contrastive predictive coding, predicting encodings of image patches. The auxiliary head shares all parameters with the classifier except its final linear layer, and the two losses are updated alternately. The paper stresses that this is not data augmentation: rotated images are never passed to the classification head, so the representation must remain object-discriminative while carrying the rotation information needed by the auxiliary task. The resulting representations are more linearly separable by object category, which is the sense in which they are 'more semantic,' and this is what transfers to detecting held-out classes as anomalous.

What would settle it

Run the Appendix D channel-wise pixel-level Gaussian baseline on the five ImageNet subsets: if it achieves average precision close to or above the multi-task classifiers on, say, the snake or car subsets, then the reported gains partly reflect low-level discrimination rather than semantic awareness.

Watch

Extended reading notes

Core claim

The paper's central claim is that out-of-distribution detection only becomes well-defined and practically meaningful when the shift is semantic with respect to a stated task context, and that classifiers trained jointly with self-supervised auxiliary objectives detect such semantic anomalies better than classification-only training. On the hold-out-class benchmarks, rotation-augmented training raises average precision on CIFAR-10 from 34.92 to 40.22 for maximum softmax probability and from 36.63 to 41.56 for ODIN; on STL-10 the corresponding gains are 21.07 to 24.41 and 22.27 to 25.14, with smaller consistent improvements across the proposed ImageNet subsets. The paper also shows that improved generalization alone does not explain the effect: randomly masking a central region improves test accuracy (96.03 to 96.27) while slightly reducing anomaly-detection average precision (34.92 to 34.41), whereas rotation augmentation improves both (96.83 and 40.22). This supports the interpretation that the auxiliary objectives improve the semantic quality of the representation, not merely its average performance.

Load-bearing premise

The benchmarks measure semantic detection only if held-out classes within each dataset are not already separable by low-level cues such as where or how the object was photographed; the paper itself notes that ringneck snakes appear in human hands and race cars on race tracks, so this fails for some ImageNet subsets.

Editorial extensions

If this is right

  • The hold-out-class protocol on CIFAR-10 and STL-10, together with the five fine-grained ImageNet subsets, offers a replacement for cross-dataset OOD benchmarks in object-recognition contexts.
  • Self-supervised auxiliary objectives such as rotation prediction and contrastive predictive coding improve average precision for both MSP and ODIN scoring, including in settings where no anomalous examples are used to tune detector hyperparameters.
  • A method that improves test accuracy can still harm anomaly detection, so semantic-aware evaluation should be reported alongside generalization when models are intended for deployment.
  • Multiple hold-out trials are preferable to a single fixed anomalous-class split, because fixed splits can reward methods that exploit one particular configuration of dataset bias.
  • Anomaly-detection performance can serve as an indirect diagnostic of how much semantic content a deep representation carries.

Reading between the lines

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

  • The paper's semantic-versus-non-semantic split suggests a natural extension the authors do not run: holding out intermediate semantic levels, for example a liger relative to lion and tiger training classes, to measure detection of compositional or fine-grained novelty rather than only novel categories.
  • If the pixel-level Gaussian baseline from Appendix D were run on the ImageNet subsets, its score would reveal how much of the remaining benchmark difficulty is semantic; the paper only reports that baseline for CIFAR-10 hold-out classes.
  • The rotation-prediction benefit may not be unique: jigsaw puzzles, colorization, or other self-supervised pretext tasks could plausibly induce the same semantic bias, and comparing them would test whether the effect is about semanticity or about the specific rotation task.
  • A practical diagnostic suggested by the masking result is to measure anomaly-detection average precision alongside accuracy on a validation set; a training change that raises accuracy but lowers detection should be suspected of exploiting low-level shortcuts.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 3 minor

Summary. The paper argues that out-of-distribution detection should be evaluated on semantic shift for a specified context rather than on cross-dataset differences driven by low-level statistics, and proposes hold-out-class benchmarks on CIFAR-10 and STL-10 plus new fine-grained ImageNet subsets. The authors show that augmenting a classifier with self-supervised auxiliary tasks, rotation prediction and contrastive predictive coding, improves anomaly detection average precision and classification accuracy. The central quantitative evidence is the CIFAR-10 improvement (MSP 34.92 to 40.22, ODIN 36.63 to 41.56) and STL-10 improvement (MSP 21.07 to 24.41, ODIN 22.27 to 25.14), with smaller and less uniform gains on the ImageNet subsets. A masking control experiment distinguishes the auxiliary-objective gains from generic regularization. The paper also reports a trivial pixel-level baseline that does poorly on the CIFAR-10 hold-out task, supporting the claim that those benchmarks isolate semantic content.

Significance. If taken at face value, the CIFAR-10 and STL-10 hold-out-class results are a solid demonstration that rotation-prediction as an auxiliary objective improves semantic anomaly detection beyond what would be expected from improved accuracy alone; the masking control is a useful check, and the critical discussion of cross-dataset OOD benchmarks is timely and constructive. The paper's proposed distinction between semantic and non-semantic shift and its advocacy of multiple hold-out trials are valuable methodological contributions. However, the ImageNet subset evidence is not validated as isolating semantic shift, and the abstract's general claim that auxiliary objectives improve semantic anomaly detection overstates what the experiments establish.

major comments (2)
  1. [Appendix A; Tables 5 and 6] The ImageNet subset benchmarks are not established as isolating semantic shift. Appendix A explicitly concedes that ringneck snakes are usually photographed in human hands and race cars on race tracks, so low-level contextual cues can separate some held-out classes. No control analogous to Appendix D's pixel-level Gaussian baseline for CIFAR-10 hold-out classes (11.17 AP) is reported for the dog, car, snake, spider, or fungus subsets. Without such a check, the averaged AP gains in Table 5 (e.g., snake MSP 18.62 to 20.23; fungus ODIN 44.59 to 46.86) could reflect improved low-level discrimination rather than improved semantic awareness. The authors should either add trivial low-level baselines for every ImageNet subset or explicitly restrict the semantic-isolation claim to the CIFAR-10/STL-10 benchmarks.
  2. [Table 5 and Appendix B] The general claim that auxiliary objectives result in improved semantic anomaly detection is not supported by the ImageNet rotation experiments. The car subset is effectively flat or negative (MSP 21.54 to 21.66, ODIN 22.49 to 22.38, accuracy 77.17 to 76.72), and the dog subset shows a decrease under ODIN (25.85 to 25.73); individual member classes such as race car, convertible, and limo decrease under both scoring methods. With only three trials, most ImageNet differences are within one standard deviation of the mean. The conclusions in Section 5.2 and the Abstract should be restricted to the CIFAR-10/STL-10 results, or the ImageNet results should be supplemented with formal significance testing and a subset-level explanation of why the effect is absent where a low-level confound is strongest.
minor comments (3)
  1. [Throughout] There are several typographical errors: 'in-distrbution' in Section 3, 'TINY-I MAGENET' in Section 3, 'Convertile' in the Figure 3 caption, 'Norweigian' in Appendix B, and 'NeuRIPS' in the references.
  2. [Section 5.1] The sentence 'λ is tuned to 0.5 for CIFAR-10, 1.0 for STL-10, and a mix of 0.5 and 1.0 for IMAGENET' is vague; the per-subset λ values used for the ImageNet experiments should be reported in a table or in Appendix B.
  3. [Section 5.2] The masking control is reported only for MSP; reporting the corresponding ODIN numbers would make the argument that the gain is not generic regularization more complete.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the auxiliary-task improvements are evaluated on held-out classes, and no anomaly-detection score is used to fit or select the method.

full rationale

No circularity found. The paper's central empirical claim—that rotation-prediction and CPC auxiliary objectives improve semantic anomaly detection—is tested on held-out classes within CIFAR-10, STL-10, and ImageNet subsets, so no anomaly-detection score enters the training objective. The only tuned hyperparameter is selected on validation classification accuracy, not on anomaly detection: the paper states, "λ is a hyper-parameter we learn by optimizing for classification accuracy on the validation set." ODIN's temperature and perturbation are fixed once across all experiments: "We fix T = 1000, ϵ = 5e-5 for all experiments, following the most common setting." The benchmark construction is not circular: the held-out classes are defined independently of the method, and the paper reports a near-chance pixel-level Gaussian baseline on CIFAR-10 hold-out classes ("achieving an average precision of a mere 11.17% across the 10 experiments"), indicating that this task is not solvable by trivial low-level statistics. The admitted ImageNet subset biases in Appendix A ("ringneck snakes are most often photographed when held in human hands, and race cars are usually pictured on race tracks") are a benchmark-validity limitation, not a circular derivation; they weaken the generality of the ImageNet results but do not make the reported improvements reduce to the auxiliary objective by construction. The only self-citation, conditional normalization from [44], is used as an implementation trick for CPC experiments and is not load-bearing for the central claim; it is a peer-reviewed, parameter-free technique and is not used to define the target metric or to force the result. Therefore no step satisfies the requirement of exhibiting Eq. X = Eq. Y by construction, and the circularity score is 0.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claims rest on lambda values tuned to validation accuracy, the normative premise that semantic shift is the practically relevant anomaly setting, the operationalization of semantic novelty as held-out classes within a dataset, and the assumption that rotation and CPC auxiliary tasks induce semantic representations. No invented entities are introduced. The free parameters are all hyperparameters, and none are fitted to anomaly-detection performance, which keeps the empirical claims from being self-fulfilling.

free parameters (3)
  • lambda_rotation_auxiliary_loss_weight = 0.5 (CIFAR-10), 1.0 (STL-10), mix of 0.5 and 1.0 (ImageNet)
    Weight of the rotation-prediction auxiliary loss in Eq. 5. Tuned to maximize validation-set classification accuracy, not anomaly detection performance, which keeps it off the target metric.
  • lambda_CPC_auxiliary_loss_weight = 10.0
    Weight of the CPC auxiliary loss; selected via coarse search over {0.1, 1.0, 10.0, 20.0} for best validation classification accuracy.
  • ODIN_temperature_and_perturbation = T = 1000, epsilon = 5e-5
    Fixed for all experiments, following the most common setting in the literature; the authors note it is unclear how to choose these without access to anomalous examples.
assumptions (5)
  • domain assumption Practical OOD detection problems are about semantic distributional shift for a specified context; non-semantic shift should be handled by robustness, not detection.
    Normative premise of Sections 1 and 2 that grounds the proposed benchmarks; if real deployments require detecting non-semantic shifts too, the benchmarks miss a core case.
  • domain assumption Hold-out-class evaluation within one dataset approximates field anomaly detection, and multiple hold-out trials average out per-split dataset bias.
    Section 2 and Appendix A argue that multiple trials reduce inadvertent advantages to bias, but this is argued, not demonstrated.
  • ad hoc to paper ImageNet hierarchy subsets are semantically coherent groups whose members share semantic content but avoid low-level dataset-bias confounds.
    Appendix A admits this is violated for some members (ringneck snakes held in hands, race cars on tracks), so low-level cues can separate classes.
  • domain assumption Rotation prediction and CPC induce representations with greater semantic awareness, which transfers to better anomaly detection.
    Section 4 hypothesis; supported only indirectly via cited linear-separability results, the correlation between accuracy and average precision gains, and the masking control experiment.
  • standard math Average precision as computed by Eq. 4 is a faithful performance measure under heavy skew.
    The paper cites interpolation pitfalls (ref 26) but uses the weighted-mean estimator as a standard approximation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Detecting semantic anomalies." pith.science (2026). https://pith.science/paper/2OXYEMWA

@misc{pith2026190804388,
  author       = {Pith},
  title        = {Pith review of: Detecting semantic anomalies},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2OXYEMWA}},
  note         = {Machine review of arXiv:1908.04388}
}
read the original abstract

We critically appraise the recent interest in out-of-distribution (OOD) detection and question the practical relevance of existing benchmarks. While the currently prevalent trend is to consider different datasets as OOD, we argue that out-distributions of practical interest are ones where the distinction is semantic in nature for a specified context, and that evaluative tasks should reflect this more closely. Assuming a context of object recognition, we recommend a set of benchmarks, motivated by practical applications. We make progress on these benchmarks by exploring a multi-task learning based approach, showing that auxiliary objectives for improved semantic awareness result in improved semantic anomaly detection, with accompanying generalization benefits.

Figures

Figures reproduced from arXiv: 1908.04388 by the authors.

Figure 1
Figure 1. Plots of costs, accuracies, and average precision for hold-out-class experiments with 3 categories each from CIFAR-10 (top) and STL-10 (bottom), using the MSP method [6]. While classification performance is not correlated with performance at anomaly detection (compare test accuracy numbers with average precision scores), the “pattern” of improvement at anomaly detection appears roughly related to generalization (com… view at source ↗
Figure 2
Figure 2. Dog (hound dog) (a) Model T (b) Race car (c) Sports car (d) Minivan (e) Ambulance (f) Cab (g) Beach wagon (h) Jeep (i) Convertile (j) Limousine [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Car 13 [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Snake (colubrid) (a) Tarantula (b) Argiope aurantia (c) Barn spider (d) Black widow (e) Garden spider (f) Wolf spider [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: Spider (a) Stinkhorn (b) Bolete (c) Hen-of-the-woods (d) Earthstar (e) Gyromitra (f) Coral fungus [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Fungus 14 [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 35 canonical work pages

  1. [1]

    Weinberger

    Chuan Guo, Geoff Pleiss, Yu Sun, and Kilian Q. Weinberger. On calibration of modern neural networks. ICML, pages 1321–1330, 2017

    work page 2017

  2. [2]

    Wichmann, and Wieland Brendel

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

    2019

  3. [3]

    Strike (with) a pose: Neural networks are easily fooled by strange poses of familiar objects

    Michael Alcorn, Qi Li, Zhitao Gong, Chengfei Wang, Long Mai, Wei shinn Ku, and Anh Nguyen. Strike (with) a pose: Neural networks are easily fooled by strange poses of familiar objects. CVPR, 2019

    work page 2019

  4. [4]

    Recognition in terra incognita

    Sara Beery, Grant Van Horn, and Pietro Perona. Recognition in terra incognita. CoRR, 2018

    work page 2018

  5. [5]

    Christiano, John Schulman, and Dan Mané

    Dario Amodei, Chris Olah, Jacob Steinhardt, Paul F. Christiano, John Schulman, and Dan Mané. Concrete problems in AI safety. CoRR, abs/1606.06565, 2016

    arXiv 2016

  6. [6]

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

    Daniel Hendrycks and Kevin Gimpel. A baseline for detecting misclassified and out-of- distribution examples in neural networks. ICLR, 2017. 9

    work page 2017

  7. [7]

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

    work page 2018

  8. [8]

    Torralba and A

    A. Torralba and A. A. Efros. Unbiased look at dataset bias. In CVPR, pages 1521–1528, June 2011

    work page 2011

Show all 45 references
  1. [9]

    A Deeper Look at Dataset Bias, pages 37–55

    Tatiana Tommasi, Novi Patricia, Barbara Caputo, and Tinne Tuytelaars. A Deeper Look at Dataset Bias, pages 37–55. 2017

    2017

  2. [10]

    Do deep generative models know what they don’t know? ICLR, 2019

    Eric Nalisnick, Akihiro Matsukawa, Yee Whye Teh, Dilan Gorur, and Balaji Lakshminarayanan. Do deep generative models know what they don’t know? ICLR, 2019

    2019

  3. [11]

    Fodor and Zenon W

    Jerry A. Fodor and Zenon W. Pylyshyn. Connectionism and cognitive architecture: A critical analysis. Cognition, 28(1):3 – 71, 1988

    1988

  4. [12]

    https : / / news

    iNaturalist. https : / / news . developer . nvidia . com / ai-app-identifies-plants-and-animals-in-seconds , accessed on 17 may 2019, 2019

    2019

  5. [13]

    Artificial intelligence in pest insect monitoring

    Peter Fedor, Jaromír Vanhara, Josef Havel, Igor Malenovsky, and Ian Spellerberg. Artificial intelligence in pest insect monitoring. Systematic Entomology, 34(2):398–400, 2009

    2009

  6. [14]

    Pitman, Anabelle W

    Marco Willi, Ross T. Pitman, Anabelle W. Cardoso, Christina Locke, Alexandra Swanson, Amy Boyer, Marten Veldthuis, and Lucy Fortson. Identifying animal species in camera trap images using deep learning and citizen science. Methods in Ecology and Evolution, 10(1):80–91, 2019

    2019

  7. [15]

    Going deeper in the automated identification of herbarium specimens

    Jose Carranza-Rojas, Herve Goeau, Pierre Bonnet, Erick Mata-Montero, and Alexis Joly. Going deeper in the automated identification of herbarium specimens. BMC Evolutionary Biology, 17(1):181, Aug 2017

    2017

  8. [16]

    Chester: A web delivered locally computed chest x-ray disease prediction system

    Joseph Paul Cohen, Paul Bertin, and Vincent Frappier. Chester: A web delivered locally computed chest x-ray disease prediction system. CoRR, abs/1901.11210, 2019

    1901 arXiv

  9. [17]

    Deep learning approach to bacterial colony classification

    Bartosz Zieli´nski, Anna Plichta, Krzysztof Misztal, Przemysław Spurek, Monika Brzychczy- Włoch, and Dorota Ocho´nska. Deep learning approach to bacterial colony classification. PLoS One, 12(9), 2017

    2017

  10. [18]

    Akcay, A

    S. Akcay, A. Atapour-Abarghouei, and T. P. Breckon. Ganomaly: Semi-supervised anomaly detection via adversarial training. ACCV, 2018

    2018

  11. [19]

    Efficient gan-based anomaly detection

    Houssam Zenati, Chuan Sheng Foo, Bruno Lecouat, Gaurav Manek, and Vijay Ramaseshan Chandrasekhar. Efficient gan-based anomaly detection. CoRR, abs/1802.06222, 2018

    2018 arXiv

  12. [20]

    Generative probabilistic novelty detection with adversarial autoencoders

    Stanislav Pidhorskyi, Ranya Almohsen, and Gianfranco Doretto. Generative probabilistic novelty detection with adversarial autoencoders. NIPS, 2018

    2018

  13. [21]

    Deep anomaly detection using geometric transformations

    Izhak Golan and Ran El-Yaniv. Deep anomaly detection using geometric transformations. NeuRIPS, 2018

    2018

  14. [22]

    Berg, and Li Fei-Fei

    Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg, and Li Fei-Fei. ImageNet Large Scale Visual Recognition Challenge. IJCV, 2015

    2015

  15. [23]

    An introduction to roc analysis

    Tom Fawcett. An introduction to roc analysis. Pattern Recognition Letters, 27(8):861 – 874, 2006

    2006

  16. [24]

    The relationship between precision-recall and roc curves

    Jesse Davis and Mark Goadrich. The relationship between precision-recall and roc curves. pages 233–240, 2006

    2006

  17. [25]

    Anand Avati, Kenneth Jung, Stephanie Harman, Lance Downing, Andrew Ng, and Nigam H. Shah. Improving palliative care with deep learning. BMC Medical Informatics and Decision Making, 18(4):122, Dec 2018

    2018

  18. [26]

    Eng, and C

    Kendrick Boyd, Kevin H. Eng, and C. David Page. Area under the precision-recall curve: Point estimates and confidence intervals. ECML-PKDD, 2013. 10

    2013

  19. [27]

    Towards open set deep networks

    Abhijit Bendale and Terrance E Boult. Towards open set deep networks. ICCV, 2016

    2016

  20. [28]

    Reducing network agnostophobia

    Akshay Raj Dhamija, Manuel Günther, and Terrance Boult. Reducing network agnostophobia. NIPS, 2018

    2018

  21. [29]

    Terrance DeVries and Graham W. Taylor. Learning confidence for out-of-distribution detection in neural networks. arXiv preprint arXiv:1802.04865, 2018

    2018 arXiv

  22. [30]

    Deep anomaly detection with outlier exposure

    Daniel Hendrycks, Mantas Mazeika, and Thomas Dietterich. Deep anomaly detection with outlier exposure. ICLR, 2019

    2019

  23. [31]

    Training confidence-calibrated classifiers for detecting out-of-distribution samples

    Kimin Lee, Honglak Lee, Kibok Lee, and Jinwoo Shin. Training confidence-calibrated classifiers for detecting out-of-distribution samples. ICLR, 2018

    2018

  24. [32]

    Out-of-distribution detection using multiple semantic label representations

    Gabi Shalev, Yossi Adi, and Joseph Keshet. Out-of-distribution detection using multiple semantic label representations. NeuRIPS, 2018

    2018

  25. [33]

    Learning fair representa- tions

    Rich Zemel, Yu Wu, Kevin Swersky, Toni Pitassi, and Cynthia Dwork. Learning fair representa- tions. ICML, 2013

    2013

  26. [34]

    Multitask learning: A knowledge-based source of inductive bias

    Rich Caruana. Multitask learning: A knowledge-based source of inductive bias. In ICML, 1993

    1993

  27. [35]

    Unsupervised visual representation learning by context prediction

    Carl Doersch, Abhinav Gupta, and Alexei A Efros. Unsupervised visual representation learning by context prediction. ICCV, 2015

    2015

  28. [36]

    Context encoders: Feature learning by inpainting

    Deepak Pathak, Philipp Krähenbühl, Jeff Donahue, Trevor Darrell, and Alexei Efros. Context encoders: Feature learning by inpainting. CVPR, 2016

    2016

  29. [37]

    Unsupervised learning of visual representations by solving jigsaw puzzles

    Mehdi Noroozi and Paolo Favaro. Unsupervised learning of visual representations by solving jigsaw puzzles. ECCV, 2016

    2016

  30. [38]

    Split-brain autoencoders: Unsupervised learning by cross-channel prediction

    Richard Zhang, Phillip Isola, and Alexei A Efros. Split-brain autoencoders: Unsupervised learning by cross-channel prediction. CVPR, 2017

    2017

  31. [39]

    Representation learning with contrastive predictive coding

    Aäron van den Oord, Yazhe Li, and Oriol Vinyals. Representation learning with contrastive predictive coding. CoRR, abs/1807.03748, 2018

    2018 arXiv

  32. [40]

    Unsupervised representation learning by predicting image rotations

    Spyros Gidaris, Praveer Singh, and Nikos Komodakis. Unsupervised representation learning by predicting image rotations. ICLR, 2018

    2018

  33. [41]

    Deep clustering for unsupervised learning of visual features

    Mathilde Caron, Piotr Bojanowski, Armand Joulin, and Matthijs Douze. Deep clustering for unsupervised learning of visual features. In ECCV, 2018

    2018

  34. [42]

    Multi-task self-supervised visual learning

    Carl Doersch and Andrew Zisserman. Multi-task self-supervised visual learning. In ICCV, 2017

    2017

  35. [43]

    Wide residual networks

    Sergey Zagoruyko and Nikos Komodakis. Wide residual networks. In BMVC, 2016

    2016

  36. [44]

    Courville

    Harm de Vries, Florian Strub, Jérémie Mary, Hugo Larochelle, Olivier Pietquin, and Aaron C. Courville. Modulating early visual processing by language. NIPS, 2017

    2017

  37. [45]

    sits inside

    Olivier J Hénaff, Ali Razavi, Carl Doersch, SM Eslami, and Aaron van den Oord. Data-efficient image recognition with contrastive predictive coding. arXiv preprint arXiv:1905.09272, 2019. 11 A Imagenet benchmarks We present details of the Imagenet-based benchmark we proposed. Fo...

    1905 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.