Harnessing Hard Mixed Samples with Decoupled Regularizer

Mixup is an efficient data augmentation approach that improves the generalization of neural networks by smoothing the decision boundary with mixed data. Recently, dynamic mixup methods have improved previous static policies effectively (e.g., linear interpolation) by maximizing target-related salient regions in mixed samples, but excessive additional time costs are not acceptable. These additional computational overheads mainly come from optimizing the mixed samples according to the mixed labels. However, we found that the extra optimizing step may be redundant because label-mismatched mixed samples are informative hard mixed samples for deep models to localize discriminative features. In this paper, we thus are not trying to propose a more complicated dynamic mixup policy but rather an efficient mixup objective function with a decoupled regularizer named Decoupled Mixup (DM). The primary effect is that DM can adaptively utilize those hard mixed samples to mine discriminative features without losing the original smoothness of mixup. As a result, DM enables static mixup methods to achieve comparable or even exceed the performance of dynamic methods without any extra computation. This also leads to an interesting objective design problem for mixup training that we need to focus on both smoothing the decision boundaries and identifying discriminative features. Extensive experiments on supervised and semi-supervised learning benchmarks across seven datasets validate the effectiveness of DM as a plug-and-play module. Source code and models are available at https://github.com/Westlake-AI/openmixup

Paper

Similar papers

Peer review

Reviewer s9eU6/10 · confidence 4/52023-06-29

Summary

This paper proposes Decoupled Softmax(Eq. 4), which is an interesting improvement to the previous mixup method, which mitigates the impact of noise in mixed samples by modifying the loss.

Strengths

The proposed idea is simple and effective. The manuscript has a high degree of completion and is rich in experiments.

Weaknesses

I see no obvious disadvantages. There are some related literatures that I think are close to the author's claim: The Benefits of Mixup for Feature Learning, It found that modifying the lambda of y in the mixup loss does not significantly affect the performance of the model. UMIX: Improving Importance Weighting for Subpopulation Shift via Uncertainty-Aware Mixup, This paper also found that appropriate modifications can be made in the mixup loss to improve the generalization of the model.

Questions

I have no other issues and the paper is written clearly and easily understood.

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

NA

Reviewer sDcS6/10 · confidence 4/52023-07-05

Summary

The authors propose a new objective function with decoupled regularizer named decoupled mixup (DM) to harness hard mixed samples and mine discriminative features adaptively. This method is available on supervised learning and semi-supervised learning. Unlike the previous approaches which propose a more complicated dynamic mixup policy with extra computation, the proposed DM can adaptively utilize those hard mixed samples to mine discriminative features without losing the original smoothness of mixup.

Strengths

1. Harnessing hard mixed samples without losing the original smoothness of mixup is an interesting idea. 2. The proposed DM enables static mixup methods to achieve comparable or even exceed the performance of dynamic methods without any extra computation. 3. The authors provide lots of experiments to demonstrate the effectiveness of the proposed method.

Weaknesses

1. It is confusing about the important notations, such as i,j, a, b. The author should follow some notational conventions. 2. Section 4.2 is abrupt with a poor description, confusing notation definitions, and low contextual relevance. 3. Equation (2) is not closely related to the context. Please explain it in detail. 4. Since Table 6 reports the experiments on transfer learning, the authors should describe that their proposed DM can be adapted to transfer learning in the related work. In addition, the title of Section 5.2 should be changed to "Transfer Learning Benchmarks". It might be more reasonable if the authors exchange Sections 5.2 and 5.3 since the authors pay more attention on semi-supervised learning. 5. It would be better if the authors move lines 164 - 175 and Figure 3 to Section 5. Figures 5 and 6 are too small in the font and the colors of the lines are hard to distinguish. 6. Some important equations are not numbered, and there should be a ',' link between the equation and the "where" in the same sentence.

Questions

Please refer to the weaknesses.

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

2 fair

Presentation

2 fair

Contribution

3 good

Limitations

The authors described potential negative societal impact.

Reviewer y8Fo7/10 · confidence 3/52023-07-07

Summary

This paper introduced a simple strategy decoupled mixup (DM) to improve the effectiveness of Mixup and its variants. Regarding the softmax result of a mixed image with a pair of classes, one class is removed from the denominator when computing the loss of the other class. Authors provided both theoretical and empirical analyses to confirm DM has the effect of increasing the confidence of the predicted classes. DM can also be applied in semi-supervised learning and multi-label classification. Extensive experiments have been conducted covering standard image classification, semi-supervised learning and semi-supervised fine-tuning.

Strengths

Strengths 1. The work is well motivated with the idea of making confident predictions for Mixup training. 2. The proposed idea is novel and can be combined with existing Mixup variants. 3. DM is proved to be effective on various tasks.

Weaknesses

1. Authors claimed the smoothness of Mixup can be preserved by DM, but I didn't see detailed discussions about this. The idea of DM seems contradict with smoothness of Mixup or label smoothing. To this end, the mechanism of DM is not totally clear. 2. Why PuzzleMix and AutoMix are not evaluated in image classification tasks?

Questions

See Weaknesses

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

N/A

Reviewer JzuZ8/10 · confidence 3/52023-07-12

Summary

The authors point out that while $\textit{dynamic}$ mixup methods are shown to be effective, they induce too much computational cost. To address this issue, they propose a $\textit{static}$ method called Decoupled Mixup, which utilizes the hard mixed samples. The authors suggest that the Softmax function will suppress the model's confidence on hard mixed samples. Based on this idea, the authors propose a decoupled mixup cross-entropy loss which uses a decoupled version of Softmax that ease the "sum-to-one" constraint of Softmax. This loss is then added in the standard mixup loss as a regularization term. Empirically, the authors show that decoupled mixup improves the top-1 accuracy performance beyond some standard mixup methods on a variety of benchmark datasets. They also show that this method can be generalized with good performance on semi-supervised learning.

Strengths

1. The idea of utilizing hard mixed samples from the perspective of Softmax function is novel and interesting. 2. The theoretical explanation of the effectiveness of the proposed algorithm is solid. 3. The experiments are conducted thoroughly on plenty of tasks, datasets and Mixup methods. 4. Experiments settings are explained in details, especially the configurations of the hyperparameters of the proposed new method DM, leaving great convenience for future practitioners trying to reproduce the work. 5. The improvement of DM beyond standard Mixup methods is empirically shown to be significant. Also, while there isn't much improvement of DM beyond dynamic Mixup, the saved computational costs are also valuable.

Weaknesses

1. The typesettings of some figures and tables are a bit too dense. 2. A few typo.

Questions

1. Line 72. Should the word "conformation" be "confirmation"? 2. Proposition 1, line 133. What does it mean by saying "$\textit{to regress corresponding}\ \lambda$" in the gradient Equation (2)? 3. Line 144, "... for mixed data point $z_{(a,b)}$". Should it be "$x_{(a,b)}$"? 4. Line 147, "$\textbf{Decoupled Softmax}$". The subtitle seems not in the right place. Probably layout error. 5. Line 144-147. From my understanding the text here is to provide the definition of Softmax function and to introduce $\sigma(\cdot)$ as its denotation. I would simply put it in Section 3.1 before Equation (1), along with the descriptions of all other notations. 6. Line 148-150. The text here is basically telling the same story as line 137-140, that Softmax suppresses the confidence of the model on hard mixed samples, the sum of whose semantic information should be more than $1$. I think they can be combined together, rather than having the idea repeated twice in one paragraph. 7. Equation (4), "$\phi(z_{(a,b)})^{i,j}$". Since it is suggested in the previous text that superscripts denote the index, I think here $j$ can be put as a subscript of $\phi$ as an indication of the function, making the expression clearer. 8. Line 153, "the decoupled Softmax makes all items associated with $\lambda$ becomes $-1$ in gradient". Though the proof of this statement should be straightforward, can you provide it in the Appendix as well, since it's mentioned that "the derivatiopn is given in the A.1" while there is only a proof of Equation 1 in A.1? 9. Line 180, "multi-classification task". Should it be "multi-lable classification task" to be more specific? 10. Line 186, "... the unlabeled data with large $\lambda$ ..." . Does it mean unlabeled data with large combination weight? i.e. the $(1-\lambda)$ as in "$\hat{x}_{(a,b)}=\lambda{x_a}+(1-\lambda)u_b$" actually? 11. Equation under line 192. The expression in RHS (particularly $z_{(a,b)}$) is not clear enough to indicate that only the labeled part is retained in $\mathcal{L}_{DM}$. 12. Line 220 "threshold $t$" and line 222 "$\xi$". The notations are used exchanged. 13. Appendix A.1, first line of the equations under line 526. Should the minus sign "$-$" here be the equal sign "$=$"? 14. The idea of raising confidence for hard mixed samples is interesting, but conventionally from the perspective of calibration, one may wish the confidence to not be too large. Will there be a contradict (or a trade-off) between leveraging hard mixed samples and improving the models' calibration performance? 15. In some tasks (datasets), mixup may not necessarily create hard mixed samples, especially when the type of data doesn't have apparent semantic information, for example points on a 2D plane. Also, when manifold intrusion occurs, the true label of a mixed sample may differs from both the labels of the pair of real samples. In these cases, DM may provide no significant improvement, or even degrade the performance. Do you have any insight or principle to determine the "level of necessity" or "effectiveness" of applying DM in a given task and dataset?

Rating

8: Strong Accept: Technically strong paper, with novel ideas, excellent impact on at least one area, or high-to-excellent impact on multiple areas, with excellent evaluation, resources, and reproducibility, and no unaddressed ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

4 excellent

Presentation

3 good

Contribution

3 good

Limitations

Not many obvious limitations. The derivation of some theory statement is not complete.

Reviewer JzuZ2023-08-15

Thank you for the response. Questions addressed.

Authorsrebuttal2023-08-19

Look forward to post-rebuttal feedback

Dear Reviewers, We would like to express our sincere gratitude for dedicating your time to reviewing our manuscript. Your insightful comments and suggestions have been instrumental in refining the quality and clarity of our work. We also thank reviewer JzuZ again for his positive response and discussion participation. We have thoroughly considered ALL feedback and carefully responded to other reviewers. We hope our responses have addressed your concerns to your satisfaction. With the improvements made and clarifications provided, we kindly hope you to raise the score for our paper, if you deem fit. If you need any clarification or have any other questions, please do not hesitate to let us know. Once again, we sincerely thank you for your invaluable contribution to our work and look forward to your post-rebuttal feedback. Best regards, Authors.

Reviewer sDcS2023-08-22

Response to Rebuttal

Thank you for your response and clarifying many details to the questions. I increase the score from 5 to 6.

Program Chairsdecision2023-09-21

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC