Informed Mixing -- Improving Open Set Recognition via Attribution-based Augmentation

Jiawen Xu; Margret Keuper; Odej Kao

REVIEW 4 major objections 6 minor 14 references

Informed Mixing -- Improving Open Set Recognition via Attribution-based Augmentation

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read GradMix masks the regions a model already knows during training, forcing it to learn more diverse features; the paper reports improved open set recognition on most standard protocols.

desk verdict A well-motivated augmentation idea whose headline OSR gain isn't supported by the main comparison (it mixes detection rules), but the controlled robustness and SSL results give it enough substance for serious review. read the letter →

arxiv 2505.12803 v1 pith:4PNJHKTI submitted 2025-05-19 cs.CV cs.LG

Jiawen Xu , Odej Kao , Margret Keuper This is my paper · ORCID

classification cs.CVcs.LG

keywords opensetrecognitiondataaugmentationattributionmapsLayerCAMcontrastivelearningfeaturediversityout-of-distributiondetectioncorruptionrobustness

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

Open set recognition asks a classifier to flag images from classes it never saw during training. The paper argues that a common failure mode is feature collapse: the model locks onto a few discriminative features of known classes and never learns the broader, less obvious features that would help tell a novel class apart. To counter this, the paper proposes GradMix, a data augmentation that uses the model's own gradient-based attribution map (LayerCAM) to find the image region it has already learned, masks that region, and replaces it with a patch from another training image. It then trains with a supervised-plus-self-supervised contrastive objective, so the model must keep working even when its preferred cues are hidden. The paper reports that this raises open set AUROC on four of six standard protocols, with the largest gain on TinyImageNet, and also improves corruption robustness and self-supervised linear-probe accuracy.

What carries the argument

The load-bearing object is LayerCAM, a gradient-based attribution method that assigns a per-location weight to each feature map channel, producing finer-grained masks than Grad-CAM; the paper aggregates the attribution maps from several ResNet layers. This map is computed from the model being trained, not a fixed pretrained network, so the mask tracks what the current model finds salient. The masked view is created by patching the most activated area with a resized sample from the same minibatch, and the augmentation loss is scaled by $\gamma^2$, the fraction of masked area. These masks turn augmentations into a self-adjusting curriculum for feature diversity.

What would settle it

Compare GradMix to a control that masks a random region of identical size and shape, with the same uniform $\gamma$ distribution for the mask side length; if the random control matches or beats GradMix's AUROC on CIFAR10 and TinyImageNet, the attribution signal is not what carries the gain.

Watch

Extended reading notes

Core claim

The central discovery is that a model's own attribution map can be recycled as a training-time curriculum. GradMix computes fine-grained LayerCAM maps from the current loss, selects the single most activated region, and masks it with a patch from another sample in the minibatch; the augmented view is fed to the self-supervised branch of a SupCon+SimCLR objective, weighted by the fraction of the image area masked ($\gamma^2$). As training proceeds and the model's saliency shifts, the masks move to new regions, so the representation is repeatedly pushed off its current focus. The paper claims this yields more diverse features, better open set detection on most protocols (with over 1% AUROC gain on TinyImageNet), and consistent gains over Mixup, CutMix, and Attentive CutMix baselines.

Load-bearing premise

The load-bearing premise is that the LayerCAM peak points to the region whose features the model has already learned, so cutting out exactly that region pushes the model toward features it would not otherwise acquire; if that map lands on background or noise, GradMix would have no advantage over random masking.

Editorial extensions

If this is right

GradMix improves open set recognition AUROC on four of the six standard protocols, including a gain of more than 1% over the previous best on TinyImageNet.
GradMix reduces average accuracy drop under common corruptions at almost every severity level on CIFAR10 and TinyImageNet.
GradMix improves linear-probe top-1 accuracy for SimCLR and MoCo pretraining on TinyImageNet and ImageNet100 across ResNet18/34/50.
Because the method uses the model's own gradients, it can in principle be dropped into other contrastive or self-supervised pipelines without changing the architecture.

Reading between the lines

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

A cleaner causal test, beyond the paper's ablations, would compare GradMix against random masking of the same area and shape; the paper's ablations compare against Mixup and CutMix, not an equal-area random-CutOut control.
The same masking curriculum could plausibly help open-set detection in object detection or segmentation, where the notion of a known class is less crisp; this is an untested extension.
GradMix's gains under corruption and in linear probing suggest attribution-guided masking may serve as a general regularizer for representation learning, independent of the OSR setting.
If attribution noise early in training is a concern, a warm-up schedule that enables GradMix only after the first epochs might improve stability; the paper does not explore schedule variants.

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.

Desk Editor's Note

read the letter

Dear [Colleague],

You asked about arXiv:2505.12803. The short version: there is a genuinely new augmentation idea here, and some solid evidence it helps robustness and self-supervised learning, but the headline claim—that GradMix improves open set recognition—is weaker than the abstract suggests. The main OSR table compares numbers from other papers using a different detection score than the one used for the GradMix row, so the reported gain over the SupCon+SimCLR backbone is not actually shown.

What is new: GradMix masks the most activated region in the image, using online LayerCAM from the current model, and weights the SimCLR loss by the masked area. That is distinct from Attentive CutMix (which patches salient regions from a pretrained model) and is reasonably motivated by the idea of forcing the model to learn features outside its current focus. The controlled experiments in Figure 4, with the same detection rule and same backbone, do show GradMix ahead of no augmentation, Mixup, and CutMix on CIFAR10 and TinyImageNet. The corruption-robustness and SSL linear-probe results also look consistent and are not confounded by the detection-rule issue. Those are real contributions, and the paper is honest that it loses to ConOSR on SVHN and CIFAR10.

The problems are mostly evaluation hygiene. First, the central incremental claim—that GradMix improves OSR over the plain SupCon+SimCLR backbone—is only supported in the main text by a cross-table comparison: Table 1 uses a Mahalanobis-distance NNO score, while Table 2 uses the new cosine k-NN score from Algorithm 1. A detection-score change alone can move AUROC by several points, so the gap is uninterpretable. The same-detector comparison in Figure 4 only covers two datasets, uses k=10 instead of the main k=3, and reports no error bars. Second, there is no code, and the five-trial averages come without standard deviations. Third, the hyperparameter reporting is inconsistent: Section 3.2 says gamma is sampled from U(0.1, 0.5), but Table 8 lists gamma=1.0 for all protocols; the lambda values for CIFAR10 and TinyImageNet appear swapped between Table 8 and Appendix F. Fourth, the layer-aggregation choice was made on the same protocols used for the final numbers, so there is some selection bias. The analysis showing broader attention (the SM metric) uses the same LayerCAM maps that were used to create the augmentations, so it is partly self-confirming.

None of this kills the idea. The mechanism is plausible and the non-OSR evidence gives it independent support. But the paper as written overclaims on OSR. I would send it to peer review—the idea deserves referee time—and ask for code, same-detector baselines in the main table, error bars, and fixed hyperparameter reporting. The OSR gain should be re-measured against the actual backbone under the same scoring rule before anyone cites it.

This is a clear case of serious thinking: the work is coherent and honest, and the flaws are fixable. Not something I would cite yet, but I'd bring it to reading group to discuss evaluation discipline in augmentation papers.

Referee Report

4 major / 6 minor

Summary. The manuscript proposes GradMix, a training-time augmentation method for open set recognition. GradMix uses LayerCAM attribution maps computed from the current model and current contrastive loss to locate the most strongly activated image region, masks that region, and replaces it with a patch from another sample in the minibatch; the augmented SimCLR view is weighted by the masked area fraction in the total loss. The method is combined with SupCon and SimCLR and evaluated with a new distance-based OSR score based on cosine similarities to the top-k neighbors of each class. Across six OSR protocols, the paper reports state-of-the-art AUROC on four of them, with a gain of just over 1% on TinyImageNet, and further reports gains in corruption robustness and in SSL linear-probe accuracy across several architectures and datasets.

Significance. If the central increment were established under controlled comparison, GradMix would be a useful and simple augmentation for OSR: it requires no extra supervision, it fits naturally into contrastive training, and the reported improvements in corruption robustness and downstream linear probing suggest a broader generalization benefit. The paper's strengths are the breadth of OSR protocols, the inclusion of several common augmentation baselines, and the evaluation across SimCLR and MoCo v1 with ResNet18, ResNet34, and ResNet50. However, the central claim is not yet demonstrated because the headline OSR comparison mixes a change of detector with the change of augmentation, the only controlled ablation is reported without variance and with a different k than the main protocol, and the hyperparameter documentation is internally inconsistent.

major comments (4)

[Table 2 vs. Table 1, Section 3.3] The paper's central incremental claim is that GradMix improves OSR over the SupCon+SSL backbone, but no same-detector no-GradMix baseline appears in the main results. The GradMix numbers in Table 2 use the new Algorithm 1 score (cosine similarities to the top-k neighbors with a normalized maximum), whereas the SupCon+SSL numbers in Table 1 use the Mahalanobis-distance NNO score described in Appendix A. A change of scoring rule alone can shift AUROC by several points, so the observed gaps (e.g., 77.48 vs. 81.92 on TinyImageNet) cannot be attributed to GradMix. Please add the SupCon+SSL model without GradMix evaluated with Algorithm 1 under the same k, gamma, lambda, and training schedule as the GradMix runs.
[Figure 4, Section 4.1] Figure 4 (left) is the only same-detector controlled comparison in the paper, but it reports no standard deviations even though Section 4.1 states that all results are averages of five trials, and it fixes k=10 whereas Table 8 reports k=3 for the main OSR protocols. Without error bars and with a different k than the headline experiments, the claimed 'over 3%' gain of GradMix over no augmentation cannot be assessed, and the figure is not directly aligned with the protocol used for Table 2. Please report mean and standard deviation for the five trials and run the augmentation ablation with the same k and loss hyperparameters as the main results.
[Section 3.2, Table 8, Appendix F] The hyperparameter documentation is internally inconsistent and prevents reproduction. Section 3.2 defines gamma as the mask side-length ratio with gamma ~ U(0.1, 0.5) and Eq. (8) weights the GradMix loss by gamma^2, but Table 8 lists gamma=1.0 for every protocol. In addition, the lambda values for CIFAR10 and TinyImageNet are reversed between Table 8 and Appendix F: Table 8 gives CIFAR10 lambda=1.2 and TinyImageNet lambda=1.0, while Appendix F states the opposite. Please provide a single consistent hyperparameter table and clarify whether the gamma column is the mask ratio, the SupCon loss weight theta, or a separate quantity.
[Section 4.1, Section 5, Figure 4 (right)] The layer-aggregation choice is selected on the test OSR AUROC of the same protocols that are later reported as final results, and the supporting SM analysis in Section 5 counts high-valued pixels in the very LayerCAM maps used to create the GradMix augmentations. Both practices make the evidence for the proposed mechanism partly self-fulfilling. A hold-out validation split for selecting the layer set, together with an independent measure of learned feature breadth (e.g., segmentation overlap or transfer to an unrelated task), would be needed to support the claim that GradMix encourages broader feature learning.

minor comments (6)

[Section 1, Section 5] There are several typos: 'coindice' should be 'coincides' in Section 1, and Section 5 refers to 'LayerGAM' instead of 'LayerCAM'.
[Algorithm 1, line 8] The similarity computation uses sim(ztest, Zc), but the loop variable is zi; this should be sim(zi, Zc) for consistency.
[Appendix D.2] The sentence 'Except for MNIST and SVHN, we aggregate layer conv3_2, conv4_2, conv5_2' conflicts with Table 8, which lists MNIST and SVHN among the protocols with reported hyperparameters; please clarify whether GradMix is used for MNIST and SVHN and which layers are aggregated for those datasets.
[Appendix B, Figure 9] The Attentive CutMix comparison uses an ImageNet-pretrained ResNet18 to compute attention maps, and the text notes this gives an unfair advantage; please state explicitly how the semantic overlap between ImageNet and CIFAR/TinyImageNet was handled or restrict the comparison to baselines that do not use pretrained feature extractors.
[Section 4.4, Table 5] Please define the overbar notation for average accuracy drop consistently and state whether the reported differences, such as 18.08 vs. 15.97 on CIFAR10, are averaged over multiple training trials or a single run.
[Section 6] The paper states that code will be published upon acceptance, but no code or standard deviations are available in the current version; given the hyperparameter inconsistencies, releasing the code or at least the exact per-protocol settings would be important for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the central OSR claim is an empirical comparison on external open-set benchmarks.

full rationale

GradMix's core claim is empirical: applying LayerCAM-guided masking to the contrastive branch improves OSR AUROC on held-out open-set splits. No equation in Sections 3.2-3.3 or in the ablations derives an OSR metric from the augmentation's own construction. The detection score (cosine similarity to top-k training features) is external to the augmentation pipeline, and the mask ratio gamma is a pre-set input rather than a parameter fitted to the OSR targets. The only potentially self-referential evidence is the Section 5 SM analysis, which measures activation breadth with the same LayerCAM maps used by GradMix; however, that analysis is presented as supporting visualization and quantification, not as the paper's OSR prediction, and the main AUROC results depend on independent open-set labels. Citations to prior OSR methods are external and are not used to define the method's outputs. The experimental concerns about cross-table detector differences and hyperparameter inconsistencies are validity and comparability issues, not circular derivations.

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

The central claim rests on a small number of hand-chosen hyperparameters (loss weights, mask size bounds, k, layer set) and on three domain assumptions: feature diversity helps OSR, LayerCAM localizes learned concepts, and SupCon+SimCLR reduces class collapse. No new mechanistic entities are introduced.

free parameters (5)

SupCon loss weight theta (labeled gamma in Table 8) = 1.0
Chosen per protocol; balances supervised and self-supervised contrastive terms in Equation 8.
SimCLR/GradMix loss weight lambda = 1.0 for MNIST/SVHN/TinyImageNet; 1.2 for CIFAR10/CIFAR+10/CIFAR+50
Chosen per protocol without reported sensitivity analysis; directly scales the GradMix term.
Mask side-length ratio range gamma ~ U(gamma_min, gamma_max) = gamma_min=0.1, gamma_max=0.5
Chosen to avoid covering the full object or exceeding image margins; no ablation on these bounds.
Number of nearest neighbors k in OSR score = 3 for main OSR table; 10 in ablations
The detection score depends on top-k summed cosine similarities; protocol-dependent choice is not justified by a sensitivity study.
LayerCAM layer set for attribution aggregation = conv3_2 + conv4_2 + conv5_2
Selected based on ablation on CIFAR10 and TinyImageNet (Figure 4, right); this is model selection on the same datasets used for final reporting.

assumptions (4)

domain assumption Higher feature diversity in representations improves open set recognition (Wang et al., 2024).
Used in Section 1 and 3 to justify the SSL+GradMix design; accepted from prior theoretical work rather than re-derived.
domain assumption LayerCAM attribution maps identify the concepts the model has already learned during training.
Core heuristic in Section 3.2; if false, GradMix does not mask learned concepts and the mechanism collapses.
domain assumption Combining SupCon and SimCLR reduces class collapse and preserves more features.
Motivated by cited works (Jing et al., Chen et al., Xue et al.) and by the authors' own Table 1; treated as a starting point.
ad hoc to paper The loss of the masked SimCLR view should be weighted by the masked area fraction gamma^2.
Introduced in Equation 8 without derivation or ablation on the weighting scheme.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Informed Mixing -- Improving Open Set Recognition via Attribution-based Augmentation." pith.science (2026). https://pith.science/paper/4PNJHKTI

@misc{pith2026250512803,
  author       = {Pith},
  title        = {Pith review of: Informed Mixing -- Improving Open Set Recognition via Attribution-based Augmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4PNJHKTI}},
  note         = {Machine review of arXiv:2505.12803}
}

read the original abstract

Open set recognition (OSR) is devised to address the problem of detecting novel classes during model inference. Even in recent vision models, this remains an open issue which is receiving increasing attention. Thereby, a crucial challenge is to learn features that are relevant for unseen categories from given data, for which these features might not be discriminative. To facilitate this process and "optimize to learn" more diverse features, we propose GradMix, a data augmentation method that dynamically leverages gradient-based attribution maps of the model during training to mask out already learned concepts. Thus GradMix encourages the model to learn a more complete set of representative features from the same data source. Extensive experiments on open set recognition, close set classification, and out-of-distribution detection reveal that our method can often outperform the state-of-the-art. GradMix can further increase model robustness to corruptions as well as downstream classification performance for self-supervised learning, indicating its benefit for model generalization.

Figures

Figures reproduced from arXiv: 2505.12803 by the authors.

**Figure 2.** Graphical illustration of GradMix. Three blocks with dashed borderlines [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 4.** Left: OSR performances of the models with different augmentation methods on CIFAR10 and TinyImageNet protocols. Clear improvements can be brought by extra data augmentations. And GradMix performs best among all augmentation methods. Right: OSR performances of models with GradMix computed using different layers and layer aggregation in ResNet18. The results indicate that different layers can produce non-negligible … view at source ↗

Figures from the paper (6 more)

**Figure 5.** Figure 5: Classification accuracy drop of the models trained with and without [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

**Figure 8.** Figure 8: Change of SM with τ . The values of GradMix models are always higher, indicating broader activated areas. to OOD problems (Ming et al., 2022; Li et al., 2024; Miyai et al., 2024; Wang et al., 2023), which is similar to OSR. However, there are very few OSR solutions rel…

**Figure 7.** Figure 7: Attribution maps visualizations using LayerCAM: [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

**Figure 9.** Figure 9: A graphical comparison between Attentive Mixup [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]

**Figure 10.** Figure 10: Attribution maps from CIFAR10 and TinyIma [PITH_FULL_IMAGE:figures/full_fig_p014_10.png]

**Figure 11.** Figure 11: Attribution maps from ImageNet100 computed [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

14 extracted references · 8 canonical work pages

[1]

To ensure fair- ness, all models were trained for 600 epochs using default augmentations for contrastive learning as specified in their respective original papers

All models use ResNet18 as the backbone, with a feature dimension of 128 for contrastive learning. To ensure fair- ness, all models were trained for 600 epochs using default augmentations for contrastive learning as specified in their respective original papers. As with the OSR experiments described in Section 4.1, the final results are averaged over five...

work page 2012
[3]

and Chan, P

Hassen, M. and Chan, P. K. Learning a neural-network- based representation for open set recognition. In Proceed- ings of the 2020 SIAM International Conference on Data Mining, pp. 154–162. SIAM,

work page 2020
[7]

Rebuffi, S.-A., Kolesnikov, A., Sperl, G., and Lampert, C. H. icarl: Incremental classifier and representation learning. In CVPR, pp. 2001–2010,

work page 2001
[9]

Understanding the effect of sparsity on neural networks robustness

Timpl, L., Entezari, R., Sedghi, H., Neyshabur, B., and Saukh, O. Understanding the effect of sparsity on neural networks robustness. arXiv preprint arXiv:2206.10915,

work page Pith review arXiv
[10]

Figure 10: Attribution maps from CIFAR10 and TinyIma- geNet datasets computed using LayerCAM: Row 1,3,5,7: model trained without GradMix; Row 2,4,6,8: model trained with GradMix

Aligned with 5, models with GradMix focus on a broader range of regions in the data. Figure 10: Attribution maps from CIFAR10 and TinyIma- geNet datasets computed using LayerCAM: Row 1,3,5,7: model trained without GradMix; Row 2,4,6,8: model trained with GradMix. Models with GradMix demonstrate larger activated areas and better object focus in data. 14 Un...

work page 2022
[128]

We list the final applied hyper-parameters in Table 8, including γ andλ in the learning objective,k in the detection method, the number of training epochs, and batch size (BS)

Except for MNIST and SVHN, we aggregate layer conv3 2, conv4 2, conv5 2 in ResNet18 to compute the attribution maps for GradMix. We list the final applied hyper-parameters in Table 8, including γ andλ in the learning objective,k in the detection method, the number of training epochs, and batch size (BS). A cosine annealing scheduler is applied to adjust t...

work page 2020
[2011]

Oord, A. v. d., Li, Y ., and Vinyals, O. Representation learn- ing with contrastive predictive coding. arXiv preprint arXiv:1807.03748,

arXiv
[2012]

and Taylor, G

DeVries, T. and Taylor, G. W. Improved regularization of convolutional neural networks with cutout.arXiv preprint arXiv:1708.04552,

arXiv

Show all 14 references

[2017]

and Lampos, V

Shu, Y . and Lampos, V . Unsupervised hard negative augmentation for contrastive learning. arXiv preprint arXiv:2401.02594,

arXiv
[2019]

N., and Lopez-Paz, D

Zhang, H., Cisse, M., Dauphin, Y . N., and Lopez-Paz, D. mixup: Beyond empirical risk minimization. arXiv preprint arXiv:1710.09412,

arXiv
[2020]

Sparsity winning twice: Better robust generalization from more efficient training

Chen, T., Zhang, Z., Wang, P., Balachandra, S., Ma, H., Wang, Z., and Wang, Z. Sparsity winning twice: Better robust generalization from more efficient training. arXiv preprint arXiv:2202.09844, 2022c. Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei, L. Imagenet...

arXiv
[2021]

Understand- ing dimensional collapse in contrastive self-supervised learning

Jing, L., Vincent, P., LeCun, Y ., and Tian, Y . Understand- ing dimensional collapse in contrastive self-supervised learning. arXiv preprint arXiv:2110.09348,

arXiv
[2022]

Atten- tive cutmix: An enhanced data augmentation approach for deep learning based image classification

Walawalkar, D., Shen, Z., Liu, Z., and Savvides, M. Atten- tive cutmix: An enhanced data augmentation approach for deep learning based image classification. arXiv preprint arXiv:2003.13048,

2003 arXiv
[2024]

and Carvalho, M

Mahdavi, A. and Carvalho, M. A survey on open set recog- nition. In 2021 IEEE Fourth International Conference on Artificial Intelligence and Knowledge Engineering (AIKE), pp. 37–44. IEEE,

2021

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

To ensure fair- ness, all models were trained for 600 epochs using default augmentations for contrastive learning as specified in their respective original papers

All models use ResNet18 as the backbone, with a feature dimension of 128 for contrastive learning. To ensure fair- ness, all models were trained for 600 epochs using default augmentations for contrastive learning as specified in their respective original papers. As with the OSR experiments described in Section 4.1, the final results are averaged over five...

work page 2012

[2] [3]

and Chan, P

Hassen, M. and Chan, P. K. Learning a neural-network- based representation for open set recognition. In Proceed- ings of the 2020 SIAM International Conference on Data Mining, pp. 154–162. SIAM,

work page 2020

[3] [7]

Rebuffi, S.-A., Kolesnikov, A., Sperl, G., and Lampert, C. H. icarl: Incremental classifier and representation learning. In CVPR, pp. 2001–2010,

work page 2001

[4] [9]

Understanding the effect of sparsity on neural networks robustness

Timpl, L., Entezari, R., Sedghi, H., Neyshabur, B., and Saukh, O. Understanding the effect of sparsity on neural networks robustness. arXiv preprint arXiv:2206.10915,

work page Pith review arXiv

[5] [10]

Figure 10: Attribution maps from CIFAR10 and TinyIma- geNet datasets computed using LayerCAM: Row 1,3,5,7: model trained without GradMix; Row 2,4,6,8: model trained with GradMix

Aligned with 5, models with GradMix focus on a broader range of regions in the data. Figure 10: Attribution maps from CIFAR10 and TinyIma- geNet datasets computed using LayerCAM: Row 1,3,5,7: model trained without GradMix; Row 2,4,6,8: model trained with GradMix. Models with GradMix demonstrate larger activated areas and better object focus in data. 14 Un...

work page 2022

[6] [128]

We list the final applied hyper-parameters in Table 8, including γ andλ in the learning objective,k in the detection method, the number of training epochs, and batch size (BS)

Except for MNIST and SVHN, we aggregate layer conv3 2, conv4 2, conv5 2 in ResNet18 to compute the attribution maps for GradMix. We list the final applied hyper-parameters in Table 8, including γ andλ in the learning objective,k in the detection method, the number of training epochs, and batch size (BS). A cosine annealing scheduler is applied to adjust t...

work page 2020

[7] [2011]

Oord, A. v. d., Li, Y ., and Vinyals, O. Representation learn- ing with contrastive predictive coding. arXiv preprint arXiv:1807.03748,

arXiv

[8] [2012]

and Taylor, G

DeVries, T. and Taylor, G. W. Improved regularization of convolutional neural networks with cutout.arXiv preprint arXiv:1708.04552,

arXiv