[{"publication":"Applied Intelligence","page":"9483-9498","abstract":[{"text":"Deep neural networks are known to be vulnerable to malicious perturbations. Current methods for improving adversarial robustness make use of either implicit or explicit regularization, with the latter is usually based on adversarial training. Randomized smoothing, the averaging of the classifier outputs over a random distribution centered in the sample, has been shown to guarantee a classifier’s performance subject to bounded perturbations of the input. In this work, we study the application of randomized smoothing to improve performance on unperturbed data and increase robustness to adversarial attacks. We propose to combine smoothing along with adversarial training and randomization approaches, and find that doing so significantly improves the resilience compared to the baseline. We examine our method’s performance on common whitebox (FGSM, PGD) and black-box (transferable attack and NAttack) attacks on CIFAR-10 and CIFAR-100, and determine that for a low number of iterations, smoothing provides a significant performance boost that persists even for perturbations with a high attack norm, . For example, under a PGD-10 attack on CIFAR-10 using Wide-ResNet28-4, we achieve 60.3% accuracy for infinity norm ∞ = 8/255 and 13.1% accuracy for ∞ = 35/255 – outperforming previous art by 3% and 6%, respectively. We achieve nearly twice the accuracy on ∞ = 35/255 and even more so for perturbations with higher infinity norm. A reference implementation of the proposed method is provided. ","lang":"eng"}],"volume":53,"scopus_import":"1","article_type":"original","day":"09","citation":{"mla":"Nemcovsky, Yaniv, et al. “Adversarial Robustness via Noise Injection in Smoothed Models.” <i>Applied Intelligence</i>, vol. 53, no. 8, Springer Nature, 2022, pp. 9483–98, doi:<a href=\"https://doi.org/10.1007/s10489-022-03423-5\">10.1007/s10489-022-03423-5</a>.","ista":"Nemcovsky Y, Zheltonozhskii E, Baskin C, Chmiel B, Bronstein AM, Mendelson A. 2022. Adversarial robustness via noise injection in smoothed models. Applied Intelligence. 53(8), 9483–9498.","ama":"Nemcovsky Y, Zheltonozhskii E, Baskin C, Chmiel B, Bronstein AM, Mendelson A. Adversarial robustness via noise injection in smoothed models. <i>Applied Intelligence</i>. 2022;53(8):9483-9498. doi:<a href=\"https://doi.org/10.1007/s10489-022-03423-5\">10.1007/s10489-022-03423-5</a>","chicago":"Nemcovsky, Yaniv, Evgenii Zheltonozhskii, Chaim Baskin, Brian Chmiel, Alex M. Bronstein, and Avi Mendelson. “Adversarial Robustness via Noise Injection in Smoothed Models.” <i>Applied Intelligence</i>. Springer Nature, 2022. <a href=\"https://doi.org/10.1007/s10489-022-03423-5\">https://doi.org/10.1007/s10489-022-03423-5</a>.","ieee":"Y. Nemcovsky, E. Zheltonozhskii, C. Baskin, B. Chmiel, A. M. Bronstein, and A. Mendelson, “Adversarial robustness via noise injection in smoothed models,” <i>Applied Intelligence</i>, vol. 53, no. 8. Springer Nature, pp. 9483–9498, 2022.","short":"Y. Nemcovsky, E. Zheltonozhskii, C. Baskin, B. Chmiel, A.M. Bronstein, A. Mendelson, Applied Intelligence 53 (2022) 9483–9498.","apa":"Nemcovsky, Y., Zheltonozhskii, E., Baskin, C., Chmiel, B., Bronstein, A. M., &#38; Mendelson, A. (2022). Adversarial robustness via noise injection in smoothed models. <i>Applied Intelligence</i>. Springer Nature. <a href=\"https://doi.org/10.1007/s10489-022-03423-5\">https://doi.org/10.1007/s10489-022-03423-5</a>"},"publication_identifier":{"issn":["0924-669X"],"eissn":["1573-7497"]},"language":[{"iso":"eng"}],"issue":"8","oa_version":"None","publisher":"Springer Nature","intvolume":"        53","author":[{"first_name":"Yaniv","last_name":"Nemcovsky","full_name":"Nemcovsky, Yaniv"},{"last_name":"Zheltonozhskii","full_name":"Zheltonozhskii, Evgenii","first_name":"Evgenii"},{"first_name":"Chaim","full_name":"Baskin, Chaim","last_name":"Baskin"},{"full_name":"Chmiel, Brian","last_name":"Chmiel","first_name":"Brian"},{"id":"58f3726e-7cba-11ef-ad8b-e6e8cb3904e6","first_name":"Alexander","orcid":"0000-0001-9699-8730","last_name":"Bronstein","full_name":"Bronstein, Alexander"},{"full_name":"Mendelson, Avi","last_name":"Mendelson","first_name":"Avi"}],"extern":"1","date_updated":"2024-10-09T11:04:54Z","title":"Adversarial robustness via noise injection in smoothed models","year":"2022","doi":"10.1007/s10489-022-03423-5","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","quality_controlled":"1","status":"public","_id":"18211","date_created":"2024-10-08T12:47:53Z","month":"08","article_processing_charge":"No","publication_status":"published","fulldoi":"https://doi.org/10.1007/s10489-022-03423-5","type":"journal_article","date_published":"2022-08-09T00:00:00Z"},{"_id":"18232","year":"2022","date_updated":"2024-10-15T07:30:00Z","title":"Intra-class low-rank regularization for supervised and semi-supervised cross-modal retrieval","abstract":[{"lang":"eng","text":"Cross-modal retrieval aims to retrieve related items across different modalities, for example, using an image query to retrieve related text. The existing deep methods ignore both the intra-modal and inter-modal intra-class low-rank structures when fusing various modalities, which decreases the retrieval performance. In this paper, two deep models (denoted as ILCMR and Semi-ILCMR) based on intra-class low-rank regularization are proposed for supervised and semi-supervised cross-modal retrieval, respectively. Specifically, ILCMR integrates the image network and text network into a unified framework to learn a common feature space by imposing three regularization terms to fuse the cross-modal data. First, to align them in the label space, we utilize semantic consistency regularization to convert the data representations to probability distributions over the classes. Second, we introduce an intra-modal low-rank regularization, which encourages the intra-class samples that originate from the same space to be more relevant in the common feature space. Third, an inter-modal low-rank regularization is applied to reduce the cross-modal discrepancy. To enable the low-rank regularization to be optimized using automatic gradients during network back-propagation, we propose the rank-r approximation and specify the explicit gradients for theoretical completeness. In addition to the three regularization terms that rely on label information incorporated by ILCMR, we propose Semi-ILCMR in the semi-supervised regime, which introduces a low-rank constraint before projecting the general representations into the common feature space. Extensive experiments on four public cross-modal datasets demonstrate the superiority of ILCMR and Semi-ILCMR over other state-of-the-art methods."}],"volume":52,"doi":"10.1007/s10489-021-02308-3","quality_controlled":"1","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","scopus_import":"1","status":"public","day":"01","article_type":"original","extern":"1","publisher":"Springer Nature","intvolume":"        52","publication":"Applied Intelligence","author":[{"full_name":"Kang, Peipei","last_name":"Kang","first_name":"Peipei"},{"full_name":"Lin, Zehang","last_name":"Lin","first_name":"Zehang"},{"first_name":"Zhenguo","full_name":"Yang, Zhenguo","last_name":"Yang"},{"full_name":"Fang, Xiaozhao","last_name":"Fang","first_name":"Xiaozhao"},{"id":"58f3726e-7cba-11ef-ad8b-e6e8cb3904e6","first_name":"Alexander","orcid":"0000-0001-9699-8730","last_name":"Bronstein","full_name":"Bronstein, Alexander"},{"last_name":"Li","full_name":"Li, Qing","first_name":"Qing"},{"first_name":"Wenyin","full_name":"Liu, Wenyin","last_name":"Liu"}],"page":"33-54","oa_version":"None","date_published":"2022-01-01T00:00:00Z","type":"journal_article","language":[{"iso":"eng"}],"fulldoi":"https://doi.org/10.1007/s10489-021-02308-3","article_processing_charge":"No","month":"01","publication_status":"published","citation":{"ama":"Kang P, Lin Z, Yang Z, et al. Intra-class low-rank regularization for supervised and semi-supervised cross-modal retrieval. <i>Applied Intelligence</i>. 2022;52:33-54. doi:<a href=\"https://doi.org/10.1007/s10489-021-02308-3\">10.1007/s10489-021-02308-3</a>","mla":"Kang, Peipei, et al. “Intra-Class Low-Rank Regularization for Supervised and Semi-Supervised Cross-Modal Retrieval.” <i>Applied Intelligence</i>, vol. 52, Springer Nature, 2022, pp. 33–54, doi:<a href=\"https://doi.org/10.1007/s10489-021-02308-3\">10.1007/s10489-021-02308-3</a>.","ista":"Kang P, Lin Z, Yang Z, Fang X, Bronstein AM, Li Q, Liu W. 2022. Intra-class low-rank regularization for supervised and semi-supervised cross-modal retrieval. Applied Intelligence. 52, 33–54.","short":"P. Kang, Z. Lin, Z. Yang, X. Fang, A.M. Bronstein, Q. Li, W. Liu, Applied Intelligence 52 (2022) 33–54.","apa":"Kang, P., Lin, Z., Yang, Z., Fang, X., Bronstein, A. M., Li, Q., &#38; Liu, W. (2022). Intra-class low-rank regularization for supervised and semi-supervised cross-modal retrieval. <i>Applied Intelligence</i>. Springer Nature. <a href=\"https://doi.org/10.1007/s10489-021-02308-3\">https://doi.org/10.1007/s10489-021-02308-3</a>","ieee":"P. Kang <i>et al.</i>, “Intra-class low-rank regularization for supervised and semi-supervised cross-modal retrieval,” <i>Applied Intelligence</i>, vol. 52. Springer Nature, pp. 33–54, 2022.","chicago":"Kang, Peipei, Zehang Lin, Zhenguo Yang, Xiaozhao Fang, Alex M. Bronstein, Qing Li, and Wenyin Liu. “Intra-Class Low-Rank Regularization for Supervised and Semi-Supervised Cross-Modal Retrieval.” <i>Applied Intelligence</i>. Springer Nature, 2022. <a href=\"https://doi.org/10.1007/s10489-021-02308-3\">https://doi.org/10.1007/s10489-021-02308-3</a>."},"date_created":"2024-10-08T12:56:38Z","publication_identifier":{"eissn":["1573-7497"],"issn":["0924-669X"]}}]