Search

Your search keyword '"Zhu, Yao"' showing total 163 results
Start Over Author "Zhu, Yao" Publisher ieee
163 results on '"Zhu, Yao"'

17. Recent Advances and Challenges of 2D Fourier Transform Computational Accelerator Using GHz Ultrasonics

Author

Chen, Daniel Ssu-Han, primary, Teo, Yong Shun, additional, Yeo, Yi Xuan, additional, Trivedi, Shyam, additional, Tan, Xing Haw Marvin, additional, Yang, Zaifeng, additional, Wai, Eva Leong Ching, additional, Wang, Yu-Shun, additional, Sharma, Jaibir, additional, Bui, Viet Phuong, additional, Png, Ching Eng, additional, Lal, Amit, additional, Zhu, Yao, additional, and Chai, Kevin Tshun Chuan, additional

Published
2023
Full Text
View/download PDF

22. Transaudio: Towards the Transferable Adversarial Audio Attack Via Learning Contextualized Perturbations

Author

Gege, Qi, Chen, Yuefeng, Mao, Xiaofeng, Zhu, Yao, Hui, Binyuan, Li, Xiaodan, Zhang, Rong, and Xue, Hui

Subjects
FOS: Computer and information sciences, Sound (cs.SD), Audio and Speech Processing (eess.AS), FOS: Electrical engineering, electronic engineering, information engineering, Computer Science - Sound, Electrical Engineering and Systems Science - Audio and Speech Processing
Abstract

In a transfer-based attack against Automatic Speech Recognition (ASR) systems, attacks are unable to access the architecture and parameters of the target model. Existing attack methods are mostly investigated in voice assistant scenarios with restricted voice commands, prohibiting their applicability to more general ASR related applications. To tackle this challenge, we propose a novel contextualized attack with deletion, insertion, and substitution adversarial behaviors, namely TransAudio, which achieves arbitrary word-level attacks based on the proposed two-stage framework. To strengthen the attack transferability, we further introduce an audio score-matching optimization strategy to regularize the training process, which mitigates adversarial example over-fitting to the surrogate model. Extensive experiments and analysis demonstrate the effectiveness of TransAudio against open-source ASR models and commercial APIs.

Published
2023

30. A-PixelHop: A Green, Robust and Explainable Fake-Image Detector

Author

Zhu, Yao, Wang, Xinyu, Chen, Hong-Shuo, Salloum, Ronald, and Kuo, C. -C. Jay

Subjects
FOS: Computer and information sciences, Computer Science - Machine Learning, Computer Vision and Pattern Recognition (cs.CV), Computer Science - Computer Vision and Pattern Recognition, Machine Learning (cs.LG)
Abstract

A novel method for detecting CNN-generated images, called Attentive PixelHop (or A-PixelHop), is proposed in this work. It has three advantages: 1) low computational complexity and a small model size, 2) high detection performance against a wide range of generative models, and 3) mathematical transparency. A-PixelHop is designed under the assumption that it is difficult to synthesize high-quality, high-frequency components in local regions. It contains four building modules: 1) selecting edge/texture blocks that contain significant high-frequency components, 2) applying multiple filter banks to them to obtain rich sets of spatial-spectral responses as features, 3) feeding features to multiple binary classifiers to obtain a set of soft decisions, 4) developing an effective ensemble scheme to fuse the soft decisions into the final decision. Experimental results show that A-PixelHop outperforms state-of-the-art methods in detecting CycleGAN-generated images. Furthermore, it can generalize well to unseen generative models and datasets.

Published
2022

41. Latency-Critical Downlink Multiple Access: A Hybrid Approach and Reliability Maximization.

Author

Yuan, Xiaopeng, Zhu, Yao, Hu, Yulin, Jiang, Hao, Shen, Chao, and Schmeink, Anke

Abstract

In this work, we study a downlink multi-user network, where a single access point (AP) is supposed to accomplish data transmissions to all users under low latency constraints. To more effectively cope with the multiple access demand, we consider a hybrid strategy for the multi-user downlink service in finite blocklength (FBL) regime, which combines broadcasting with time-division multiple access (TDMA). In the hybrid strategy, the users are first clustered into different groups. Different groups are served in a TDMA manner with dedicated time slots, while users within each group are served together via a broadcasting signal from the AP. By taking into account the fairness of transmission reliability among all users, we formulate a problem minimizing the maximum error probability among users via jointly determining the user grouping and allocating blocklength among all groups. To address the complicated non-convex problem, we first characterize the optimal blocklength allocation under each given grouping decision, which leads to an optimal closed-form allocation solution via solving an equation system. Based on the characterized features, we are enabled to efficiently distill out the optimal grouping from all possible groupings, which forms the efficient optimal solution for the optimal joint design. Afterwards, aiming at a complexity reduction, we further propose a low-complexity iterative solution, in which the grouping is iteratively improved via the introduced operations until a convergence to a suboptimum. Finally, via simulations, we validate the proposed solutions and reveal the close optimality of the iterative solution. In addition, the hybrid strategy has shown a significant reliability advantage in comparison to pure broadcasting or TDMA, and this performance advantage becomes further enlarged in case of more users. [ABSTRACT FROM AUTHOR]

Published
2022
Full Text
View/download PDF

49. Energy Minimization of Mobile Edge Computing Networks With HARQ in the Finite Blocklength Regime.

Author

Zhu, Yao, Hu, Yulin, Schmeink, Anke, and Gross, James

Abstract

We consider a mobile edge computing (MEC) network supporting low-latency, critical offloading workloads. The task offloading from the user to the server is operated under a truncated Hybrid Automatic Repeat reQuest (HARQ) process, i.e., we consider finite retransmission attempts. Both the HARQ type-I and type-II schemes are studied. For each scheme, we first characterize the total error probability and the total energy cost, while the impact of finite blocklength (FBL) on the stochastic retransmission behavior is considered. Following the characterizations, we are interested in optimal frameworks for each considered HARQ type, where the number of potential retransmission attempts is optimized together with the duration of each transmission, while the CPU frequency at the edge node is adjusted via voltage scaling. The objective is to minimize the total energy cost with error probability threshold. We show that the resulting stochastic optimization problems can be solved by means of convex optimization. We furthermore demonstrate that sharp minima exist among the energy consumption, underlying the importance of near-optimal parameter choice in the studied scenarios. Our results underline the importance of trading off communication and computational characteristics in delay-critical MEC setups with FBL codes. [ABSTRACT FROM AUTHOR]

Published
2022
Full Text
View/download PDF

50. Toward Understanding and Boosting Adversarial Transferability From a Distribution Perspective.

Author

Zhu, Yao, Chen, Yuefeng, Li, Xiaodan, Chen, Kejiang, He, Yuan, Tian, Xiang, Zheng, Bolun, Chen, Yaowu, and Huang, Qingming

Subjects
*ARTIFICIAL neural networks, *DATA distribution
Abstract

Transferable adversarial attacks against Deep neural networks (DNNs) have received broad attention in recent years. An adversarial example can be crafted by a surrogate model and then attack the unknown target model successfully, which brings a severe threat to DNNs. The exact underlying reasons for the transferability are still not completely understood. Previous work mostly explores the causes from the model perspective, e.g., decision boundary, model architecture, and model capacity. Here, we investigate the transferability from the data distribution perspective and hypothesize that pushing the image away from its original distribution can enhance the adversarial transferability. To be specific, moving the image out of its original distribution makes different models hardly classify the image correctly, which benefits the untargeted attack, and dragging the image into the target distribution misleads the models to classify the image as the target class, which benefits the targeted attack. Towards this end, we propose a novel method that crafts adversarial examples by manipulating the distribution of the image. We conduct comprehensive transferable attacks against multiple DNNs to demonstrate the effectiveness of the proposed method. Our method can significantly improve the transferability of the crafted attacks and achieves state-of-the-art performance in both untargeted and targeted scenarios, surpassing the previous best method by up to 40% in some cases. In summary, our work provides new insight into studying adversarial transferability and provides a strong counterpart for future research on adversarial defense. [ABSTRACT FROM AUTHOR]

Published
2022
Full Text
View/download PDF

Catalog

Books, media, physical & digital resources
See catalog results