October 27th, 2022 
Daniela Lopez de Luise ;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px” target=”_blank”>https://forms.gle/9JCc6NBgM1x2kZK6A
or via the website of the journal at:
https://jivp-eurasipjournals.springeropen.com/
Contact: Esinu Abadjivor <esinu.abadjivor@springernature.com>
Title: Detect and Defense Against Adversarial Examples in Deep Learning using Natural Scene Statistics and Adaptive Denoising
Abstract: Despite the enormous performance of Deep Neural Networks (DNNs), recent studies have shown their vulnerability to Adversarial Examples (AEs), i.e., carefully perturbed inputs designed to fool the targeted DNN. The literature is rich with many effective attacks to craft such AEs. Meanwhile, many defense strategies have been developed to mitigate this vulnerability. However, this latter showed their effectiveness against specific attacks and does not generalize well to different attacks. This talk presents a framework for defending the DNN classifier against adversarial samples. The proposed method includes a separate detector and a denoising block. The detector aims to detect AEs by characterizing them through natural scene statistics (NSS), where we demonstrate that the presence of adversarial perturbations alters these statistical features. The denoiser is based on Block Matching 3D (BM3D) filter fed by a threshold estimated by a Convolutional Neural Network (CNN) to project back the samples detected as AEs into their data manifold. We conduct a complete evaluation on three standard datasets: MNIST, CIFAR-10, and Tiny-ImageNet, compared with state-of-the-art defenses. Our experimental results have shown that the proposed detector achieves a high detection accuracy while providing a low false positive accuracy. Additionally, we outperform the state-of-the-art-defense techniques by improving the robustness of DNN’s.
Speaker Bio: Wassim Hamidouche is a lead researcher at the Technology Innovation Institute (TII), Abu Dhabi, UAE. He was an associate professor with INSA Rennes and a member of the Institute of Electronics and Telecommunications of Rennes (IETR), UMR CNRS 6164, and IRT Research Institute. He received his Ph.D. in signal and image processing from the University of Poitiers, France, in 2010. From 2011 to 2012, he was a research engineer with the Canon Research Centre in Rennes, France. He is the co-authors of more than 160+ papers published in image processing and computer vision. His research interests include video coding, the design of software and hardware circuits and systems for video coding standards, image quality assessment, and multimedia security.
Webinar videos are available online at https://vimeo.com/showcase/8005816.
or via the website of the journal at:
https://jivp-eurasipjournals.springeropen.com/
Contact: Esinu Abadjivor <esinu.abadjivor@springernature.com>
Title: Detect and Defense Against Adversarial Examples in Deep Learning using Natural Scene Statistics and Adaptive Denoising
Abstract: Despite the enormous performance of Deep Neural Networks (DNNs), recent studies have shown their vulnerability to Adversarial Examples (AEs), i.e., carefully perturbed inputs designed to fool the targeted DNN. The literature is rich with many effective attacks to craft such AEs. Meanwhile, many defense strategies have been developed to mitigate this vulnerability. However, this latter showed their effectiveness against specific attacks and does not generalize well to different attacks. This talk presents a framework for defending the DNN classifier against adversarial samples. The proposed method includes a separate detector and a denoising block. The detector aims to detect AEs by characterizing them through natural scene statistics (NSS), where we demonstrate that the presence of adversarial perturbations alters these statistical features. The denoiser is based on Block Matching 3D (BM3D) filter fed by a threshold estimated by a Convolutional Neural Network (CNN) to project back the samples detected as AEs into their data manifold. We conduct a complete evaluation on three standard datasets: MNIST, CIFAR-10, and Tiny-ImageNet, compared with state-of-the-art defenses. Our experimental results have shown that the proposed detector achieves a high detection accuracy while providing a low false positive accuracy. Additionally, we outperform the state-of-the-art-defense techniques by improving the robustness of DNN’s.
Speaker Bio: Wassim Hamidouche is a lead researcher at the Technology Innovation Institute (TII), Abu Dhabi, UAE. He was an associate professor with INSA Rennes and a member of the Institute of Electronics and Telecommunications of Rennes (IETR), UMR CNRS 6164, and IRT Research Institute. He received his Ph.D. in signal and image processing from the University of Poitiers, France, in 2010. From 2011 to 2012, he was a research engineer with the Canon Research Centre in Rennes, France. He is the co-authors of more than 160+ papers published in image processing and computer vision. His research interests include video coding, the design of software and hardware circuits and systems for video coding standards, image quality assessment, and multimedia security.
Webinar videos are available online at https://vimeo.com/showcase/8005816.
image004.jpg@01D6FBAE.3C1BBB60” width=”435″ height=”227″ border=”0″>
