Filters








94,179 Hits in 3.3 sec

Robust Learning with Frequency Domain Regularization [article]

Weiyu Guo, Yidong Ouyang
2020 arXiv   pre-print
in transfer learning scenario without fine-tune.  ...  In this paper, we introduce a new regularization method by constraining the frequency spectra of the filter of the model.  ...  the model trained with our regularization is more robust on unseen data.  ... 
arXiv:2007.03244v1 fatcat:qv7wlm5dr5hyxe2duphienrxgy

Local Convolutions Cause an Implicit Bias towards High Frequency Adversarial Examples [article]

Josue Ortega Caro, Yilong Ju, Ryan Pyle, Sourav Dey, Wieland Brendel, Fabio Anselmi, Ankit Patel
2021 arXiv   pre-print
and frequency domains.  ...  The explanation for the latter involves the Fourier Uncertainty Principle: a spatially-limited (local in the space domain) filter cannot also be frequency-limited (local in the frequency domain).  ...  regularizer in the Fourier domain.  ... 
arXiv:2006.11440v4 fatcat:l5kykgnqlngyrgodgqlcxxwrne

Absum: Simple Regularization Method for Reducing Structural Sensitivity of Convolutional Neural Networks [article]

Sekitoshi Kanai, Yasutoshi Ida, Yasuhiro Fujiwara, Masanori Yamada, Shuichi Adachi
2019 arXiv   pre-print
Furthermore, we reveal that robust CNNs with Absum are more robust against transferred attacks due to decreasing the common sensitivity and against high-frequency noise than standard regularization methods  ...  We also reveal that Absum can improve robustness against gradient-based attacks (projected gradient descent) when used with adversarial training.  ...  High-Frequency Attack To evaluate robustness in the frequency domain, we used High-Frequency attacks.  ... 
arXiv:1909.08830v1 fatcat:bvc2odu7fvhp7kb7ivftgeve44

Feature Stylization and Domain-aware Contrastive Learning for Domain Generalization [article]

Seogkyu Jeon, Kibeom Hong, Pilhyeon Lee, Jewook Lee, Hyeran Byun
2021 arXiv   pre-print
Afterward, we stylize the low frequency components with the novel domain styles sampled from the manipulated statistics, while preserving the shape cues in high frequency ones.  ...  Domain generalization aims to enhance the model robustness against domain shift without accessing the target domain.  ...  By incorporating them in the training, our model is allowed to learn robust representation against domain shift.  ... 
arXiv:2108.08596v1 fatcat:lpnusrrtdvemrbkq55szp6vu6u

Bridging the Gap Between Adversarial Robustness and Optimization Bias [article]

Fartash Faghri, Sven Gowal, Cristina Vasconcelos, David J. Fleet, Fabian Pedregosa, Nicolas Le Roux
2021 arXiv   pre-print
To illustrate these findings we design a novel Fourier-ℓ_∞ attack that finds adversarial examples with controllable frequencies.  ...  We demonstrate that the choice of optimizer, neural network architecture, and regularizer significantly affect the adversarial robustness of linear neural networks, providing guarantees without the need  ...  We observe that high frequency Fourier-∞ attacks succeed more easily with smaller perturbations compared with low frequency attacks.  ... 
arXiv:2102.08868v2 fatcat:qhty2vogyvdmzaza45yr2xpq4y

Absum: Simple Regularization Method for Reducing Structural Sensitivity of Convolutional Neural Networks

Sekitoshi Kanai, Yasutoshi Ida, Yasuhiro Fujiwara, Masanori Yamada, Shuichi Adachi
2020 PROCEEDINGS OF THE THIRTIETH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE AND THE TWENTY-EIGHTH INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE CONFERENCE  
Furthermore, we reveal that robust CNNs with Absum are more robust against transferred attacks due to decreasing the common sensitivity and against high-frequency noise than standard regularization methods  ...  We also reveal that Absum can improve robustness against gradient-based attacks (projected gradient descent) when used with adversarial training.  ...  High-Frequency Attack To evaluate robustness in the frequency domain, we used High-Frequency attacks.  ... 
doi:10.1609/aaai.v34i04.5865 fatcat:twjrzqwiurevdalmjlcckwfxhm

Improving robustness against common corruptions with frequency biased models [article]

Tonmoy Saikia, Cordelia Schmid, Thomas Brox
2021 arXiv   pre-print
Moreover, we propose a new regularization scheme that minimizes the total variation (TV) of convolution feature-maps to increase high-frequency robustness.  ...  In this paper, we introduce a mixture of two expert models specializing in high and low-frequency robustness, respectively.  ...  Affairs and Energy within the project "KI Delta Learning -Development of methods and tools for the efficient expansion and transformation of existing AI modules of autonomous vehicles to new domains".  ... 
arXiv:2103.16241v1 fatcat:httlzfkckbgdjpv5mb35qchjmy

Impact of Spatial Frequency Based Constraints on Adversarial Robustness [article]

Rémi Bernhard, Pierre-Alain Moellic, Martial Mermillod, Yannick Bourrier, Romain Cohendet, Miguel Solinas, Marina Reyboz
2021 arXiv   pre-print
In this paper, we investigate the robustness to adversarial perturbations of models enforced during training to leverage information corresponding to different spatial frequency ranges.  ...  Indeed, depending on the data set, the same constraint may results in very different level of robustness (up to 0.41 adversarial accuracy difference).  ...  Is frequency-based regularization compatible with adversarial training ?  ... 
arXiv:2104.12679v2 fatcat:yy5bucbjm5f5hlqg4zg2sjxpii

Graph Signal Processing for Geometric Data and Beyond: Theory and Applications [article]

Wei Hu, Jiahao Pang, Xianming Liu, Dong Tian, Chia-Wen Lin, Anthony Vetro
2021 arXiv   pre-print
We conclude with a brief discussion of open problems and challenges.  ...  image/video processing methodologies are limited, while Graph Signal Processing (GSP) -- a fast-developing field in the signal processing community -- enables processing signals that reside on irregular domains  ...  Enhancing Robustness and Generalizability with GSP 1) Robustness: "Robustness" of a deep learning network may refer to 1) robustness to noisy data or labels; 2) robustness to incomplete data; 3) robustness  ... 
arXiv:2008.01918v3 fatcat:54ankltzznerpo5t5p3lkezvzu

Spectral Leakage and Rethinking the Kernel Size in CNNs [article]

Nergis Tomen, Jan van Gemert
2021 arXiv   pre-print
Convolutional layers in CNNs implement linear filters which decompose the input into different frequency bands.  ...  Finally, we show that CNNs employing the Hamming window display increased robustness against various adversarial attacks.  ...  a convolution with a sinc function in frequency domain [41] .  ... 
arXiv:2101.10143v2 fatcat:jnc7fllxsjhm7igllfudbcco3y

A Technique to Preserve Edge Information in Single Image Super Resolution

Amisha J. Shah, Suryakant B. Gupta
2016 Procedia Computer Science  
Furthermore the smoothness of smooth edges (gradual transition in intensity) is preserved by using soft edge smoothness prior as a regularizing parameter.  ...  The proposed novel approach potted the information around curvature edges and hard edges (abrupt transition in intensity) using Non Sub-Sampled Contourlet Transform (NSCT) based learning process.  ...  An adaptive regularization method is fused with an adaptive sparse domain selection to obtain a HR image 5 .  ... 
doi:10.1016/j.procs.2016.05.186 fatcat:u6vlnhrpmrhtvgdbyrincyyvje

Subspace learning with frequency regularizer: Its application to face recognition

Zhen Lei, Dong Yi, Xiangsheng Huang, Stan Z. Li
2015 2015 International Conference on Biometrics (ICB)  
Two representative supervised subspace methods with frequency regularizer, FR-LDA and FR-SR are introduced and further applied to face recognition problem.  ...  Traditional regularizers are usually designed in spatial domain, which usually make the projection smooth.  ...  Section 3 introduces two subspace learning methods incorporated with frequency regularizer.  ... 
doi:10.1109/icb.2015.7139113 dblp:conf/icb/LeiYHL15 fatcat:w736uaq4vndr5auv3egb2sbzuu

RDA: Robust Domain Adaptation via Fourier Adversarial Attacking [article]

Jiaxing Huang, Dayan Guan, Aoran Xiao, Shijian Lu
2021 arXiv   pre-print
With FAA-generated samples, the training can continue the 'random walk' and drift into an area with a flat loss landscape, leading to more robust domain adaptation.  ...  supervised learning) as the supervised source loss has clear domain gap and the unsupervised target loss is often noisy due to the lack of annotations.  ...  Figure 1 . 1 Our robust domain adaptation alleviates overfitting effectively: Both supervised learning with source data (row 1) and unsupervised learning with target data (in rows 2 and 3 for adversarial  ... 
arXiv:2106.02874v3 fatcat:w5gzhaxuprbrtesxmrxfrilwm4

Research and Application of Regularized Sparse Filtering Model for Intelligent Fault Diagnosis Under Large Speed Fluctuation

Baokun Han, Guowei Zhang, Jinrui Wang, Xiaoyu Wang, Sixiang Jia, Jingtao He
2020 IEEE Access  
The frequency domain signals under large speed fluctuation are directly input to regularized SF for feature extraction, and softmax regression is used as a classifier for fault type identification.  ...  As an effective unsupervised learning method, sparse filtering (SF) has been successfully used in intelligent fault diagnosis.  ...  So, in this paper, regularized SF models are proposed to deal with the frequency-domain signal under large speed fluctuation.  ... 
doi:10.1109/access.2020.2975531 fatcat:wiglwfodvzhexdt63gfltkg7ty

A Novel Domain Adaptation-Based Intelligent Fault Diagnosis Model to Handle Sample Class Imbalanced Problem

Zhongwei Zhang, Mingyu Shao, Liping Wang, Sujuan Shao, Chicheng Ma
2021 Sensors  
is combined with instance reweighting to simultaneously explore the intrinsic manifold structure and remove irrelevant source-domain samples adaptively. (3) The ℓ2-norm regularization is applied as the  ...  The gear and rolling bearing datasets with class imbalanced samples are applied to validate the reliability of MRMI.  ...  Then, the time-domain samples with 1200 sample lengths are converted to 600 length samples in the frequency domain.  ... 
doi:10.3390/s21103382 pmid:34066271 fatcat:k5ptp5az6negjl7xwzu6ad3qqq
« Previous Showing results 1 — 15 out of 94,179 results