Score-Based Generative Classifiers

Zimmermann, R.; Schott, Lukas; Yang, Song; Dunn, Benjamin; Klindt, David A.

doi:10.48550/arxiv.2110.00473

Cited by 6 publications

(8 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This means that the local behavior of the discriminative classifier is determined by the entire set of classes, while our approach is purely local. In contrast, in score-based generative classification [74,95] each class has a generator, that provides the closest member for a given input. The class is then determined based on the best match.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Preventing Overfitting by Training Derivatives

Avrutskiy

2019

Advances in Intelligent Systems and Computing

View full text Add to dashboard Cite

Derivative training is a well-known method to improve the accuracy of neural networks. In the forward pass, not only the output values are computed, but also their derivatives, and their deviations from the target derivatives are included in the cost function, which is minimized with respect to the weights by a gradient-based algorithm. So far, this method has been implemented for relatively low-dimensional tasks. In this study, we apply the approach to the problem of image analysis. We consider the task of reconstructing the vertices of a cube based on its image. By training the derivatives with respect to the 6 degrees of freedom of the cube, we obtain 25 times more accurate results for noiseless inputs. The derivatives also provide important insights into the robustness problem, which is currently understood in terms of two types of network vulnerabilities. The first type is small perturbations that dramatically change the output, and the second type is substantial image changes that the network erroneously ignores. They are currently considered as conflicting goals, since conventional training methods produce a trade-off. The first type can be analyzed via the gradient of the network, but the second type requires human evaluation of the inputs, which is an oracle substitute. For the task at hand, the nearest neighbor oracle can be defined, and the knowledge of derivatives allows it to be expanded into Taylor series. This allows to perform the first-order robustness analysis that unifies both types of vulnerabilities, and to implement robust training that eliminates any trade-offs, so that accuracy and robustness are limited only by network capacity.

show abstract

Section: Discussionmentioning

confidence: 99%

“…Since each generator is described by a single manifold, the non-locality is avoided. So far, generative classifiers are not more robust [95,20] than discriminative models. Finally, studying a relatively small number of objects from different angles is more consistent with human development than learning on thousands of individual images.…”

Section: Discussionmentioning

confidence: 99%

Preventing Overfitting by Training Derivatives

Avrutskiy

2019

Advances in Intelligent Systems and Computing

View full text Add to dashboard Cite

show abstract

“…diffusion models have become a powerful family of deep generation models [58], [59], with record breaking performance in many areas [60], including image generation [61], [30], [62], [63], image inpainting [64], image super-resolution [65], [66], [67], and image-to-image translation [68], [69], [70]. In addition, the feature representations learned from the diffusion models are also found to be very useful in discriminative tasks, including image classification [71], image segmentation [72], [73] and object detection [74]. A diffusion models is a deep generative model with two processes, namely the forward process and the reverse process [75].…”

Section: B Diffusion Modelsmentioning

confidence: 99%

Dif-Fusion: Towards High Color Fidelity in Infrared and Visible Image Fusion with Diffusion Models

Yue¹,

Fang²,

Xia³

et al. 2023

Preprint

View full text Add to dashboard Cite

Color plays an important role in human visual perception, reflecting the spectrum of objects. However, the existing infrared and visible image fusion methods rarely explore how to handle multi-spectral/channel data directly and achieve high color fidelity. This paper addresses the above issue by proposing a novel method with diffusion models, termed as Dif-Fusion, to generate the distribution of the multi-channel input data, which increases the ability of multi-source information aggregation and the fidelity of colors. In specific, instead of converting multi-channel images into single-channel data in existing fusion methods, we create the multi-channel data distribution with a denoising network in a latent space with forward and reverse diffusion process. Then, we use the denoising network to extract the multi-channel diffusion features with both visible and infrared information. Finally, we feed the multi-channel diffusion features to the multi-channel fusion module to directly generate the three-channel fused image. To retain the texture and intensity information, we propose multi-channel gradient loss and intensity loss. Along with the current evaluation metrics for measuring texture and intensity fidelity, we introduce a new evaluation metric to quantify color fidelity. Extensive experiments indicate that our method is more effective than other state-of-the-art image fusion methods, especially in color fidelity. The source code will be publicly available at https://github.com/GeoVectorMatrix/Dif-Fusion.

show abstract

“…Pacchiardi and Dutta [7] used score matching for training a neural conditional exponential family to approximate the ABC likelihood, and applied it in MCMC sampling for intractable distributions and to large-dimensional time-series model. Generative models have been used as adversarially robust classifiers for complex datasets, particularly in the image classification domain [29]. Zimmermann et al [29] investigated score-based generative classification of natural images, and found marginal advantage over discriminative classifiers in terms of adversarial robustness, yet it provides a different approach to classification.…”

Section: Future Workmentioning

confidence: 99%

“…Generative models have been used as adversarially robust classifiers for complex datasets, particularly in the image classification domain [29]. Zimmermann et al [29] investigated score-based generative classification of natural images, and found marginal advantage over discriminative classifiers in terms of adversarial robustness, yet it provides a different approach to classification.…”

Section: Future Workmentioning

confidence: 99%

Classification via score-based generative modelling

Huang¹

2022

Preprint

View full text Add to dashboard Cite

In this work, we investigated the application of score-based gradient learning in discriminative and generative classification settings. Score function can be used to characterize data distribution as an alternative to density. It can be efficiently learned via score matching, and used to flexibly generate credible samples to enhance discriminative classification quality, to recover density and to build generative classifiers. We analysed the decision theories involving score-based representations, and performed experiments on simulated and real-world datasets, demonstrating its effectiveness in achieving and improving binary classification performance, and robustness to perturbations, particularly in high dimensions and imbalanced situations.

show abstract

Score-Based Generative Classifiers

Cited by 6 publications

References 11 publications

Preventing Overfitting by Training Derivatives

Preventing Overfitting by Training Derivatives

Dif-Fusion: Towards High Color Fidelity in Infrared and Visible Image Fusion with Diffusion Models

Classification via score-based generative modelling

Contact Info

Product

Resources

About