Filters








763 Hits in 2.8 sec

Real-time, Universal, and Robust Adversarial Attacks Against Speaker Recognition Systems [article]

Yi Xie, Cong Shi, Zhuohang Li, Jian Liu, Yingying Chen, Bo Yuan
2020 arXiv   pre-print
In this paper, we propose the first real-time, universal, and robust adversarial attack against the state-of-the-art deep neural network (DNN) based speaker recognition system.  ...  In addition, we improve the robustness of our attack by modeling the sound distortions caused by the physical over-the-air propagation through estimating room impulse response (RIR).  ...  Thus, the generated adversarial perturbation needs to be robust enough to remain effective under this kind of real-world distortions. Threat Model.  ... 
arXiv:2003.02301v2 fatcat:vzv2zftbtrhuxnwjcnxx4ymmty

Adversarial Attacks and Defenses in Deep Learning: from a Perspective of Cybersecurity

Shuai Zhou, Chi Liu, Dayong Ye, Tianqing Zhu, Wanlei Zhou, Philip S. Yu
2022 ACM Computing Surveys  
Further, it is difficult to evaluate the real threat of adversarial attacks or the robustness of a deep learning model, as there are no standard evaluation methods.  ...  Hence, with this paper, we review the literature to date. Additionally, we attempt to offer the first analysis framework for a systematic understanding of adversarial attacks.  ...  Threat models against DNN To classify these attacks, we have speciied the threat model against DNN by introducing the critical components of the model attacks.  ... 
doi:10.1145/3547330 fatcat:d3x3oitysvb73ado5kuaqakgtu

Moiré Attack (MA): A New Potential Risk of Screen Photos [article]

Dantong Niu, Ruohao Guo, Yisen Wang
2021 arXiv   pre-print
In this paper, we find a special phenomenon in digital image processing, the moiré effect, that could cause unnoticed security threats to DNNs.  ...  attack with the noise budget ϵ=4), high transferability rate across different models, and high robustness under various defenses.  ...  We would like to thank Hanyue Lou at Peking University in the discussion at moiré phenomenon in daily life.  ... 
arXiv:2110.10444v1 fatcat:76ufml5xyngohcmpmuj7jjzyb4

Stealthy Attack on Algorithmic-Protected DNNs via Smart Bit Flipping [article]

Behnam Ghavami, Seyd Movi, Zhenman Fang, Lesley Shannon
2021 arXiv   pre-print
To improve the robustness of DNNs, some algorithmic-based countermeasures against adversarial examples have been introduced thereafter.  ...  In this paper, we propose a new type of stealthy attack on protected DNNs to circumvent the algorithmic defenses: via smart bit flipping in DNN weights, we can reserve the classification accuracy for clean  ...  To quantify the robustness of a classifier f , we can find the IV.  ... 
arXiv:2112.13162v1 fatcat:jf6up5vonrbhvipf2sknpczdpq

Are Malware Detection Classifiers Adversarially Vulnerable to Actor-Critic based Evasion Attacks?

Hemant Rathore, Sujay C Sharma, Sanjay K. Sahay, Mohit Sewak
2022 EAI Endorsed Transactions on Scalable Information Systems  
However, the robustness of these models against well-crafted adversarial samples is not well investigated.  ...  Thus malware detection models should be investigated for vulnerabilities and mitigated to enhance their overall forensic knowledge and adversarial robustness.  ...  The ability to simulate real-world threat scenarios and develop mitigation strategies for the same makes threat modeling a vital exercise for any system.  ... 
doi:10.4108/eai.31-5-2022.174087 fatcat:42jftpdh35db7p4do6gnnlygny

Mind the Gap! A Study on the Transferability of Virtual vs Physical-world Testing of Autonomous Driving Systems [article]

Andrea Stocco, Brian Pulfer, Paolo Tonella
2021 arXiv   pre-print
real-world vehicle.  ...  world, threatening the potential of existing testing solutions when applied to physical SDCs.  ...  The DNN architecture is driving in the real world?  ... 
arXiv:2112.11255v1 fatcat:aia44y5s2vamnabgnofo3lxo2a

Realizable Universal Adversarial Perturbations for Malware [article]

Raphael Labaca-Castro, Luis Muñoz-González, Feargus Pendlebury, Gabi Dreo Rodosek, Fabio Pierazzi, Lorenzo Cavallaro
2022 arXiv   pre-print
Machine learning classifiers are vulnerable to adversarial examples -- input-specific perturbations that manipulate models' output.  ...  Our experiments limit the effectiveness of a white box Android evasion attack to ~20% at the cost of ~3% TPR at 1% FPR.  ...  Acknowledgments This research has been partially supported by the EC H2020 Project CONCORDIA (GA 830927) and the UK EP/L022710/2 and EP/P009301/1 EPSRC research grants.  ... 
arXiv:2102.06747v2 fatcat:2tlsyq3ojbdyviumrbvwzm7ipu

EnnCore: End-to-End Conceptual Guarding of Neural Architectures

Edoardo Manino, Danilo Carvalho, Yi Dong, Julia Rozanova, Xidan Song, Mustafa A. Mustafa, André Freitas, Gavin Brown, Mikel Lujan, Xiaowei Huang, Lucas C. Cordeiro
2022 AAAI Conference on Artificial Intelligence  
The EnnCore project addresses the fundamental security problem of guaranteeing safety, transparency, and robustness in neural-based architectures.  ...  In this respect, EnnCore will pioneer the dialogue between contemporary explainable neural models and full-stack neural software verification.  ...  Acknowledgment The work is funded by EPSRC grant EP/T026995/1 entitled "EnnCore: End-to-End Conceptual Guarding of Neural Architectures" under Security for all in an AI enabled society. Prof.  ... 
dblp:conf/aaai/ManinoCDRSMFBL022 fatcat:weaswjcuwjeslhv443vrgouywe

Adversarial Attack on Attackers: Post-Process to Mitigate Black-Box Score-Based Query Attacks [article]

Sizhe Chen, Zhehao Huang, Qinghua Tao, Yingwen Wu, Cihang Xie, Xiaolin Huang
2022 arXiv   pre-print
The score-based query attacks (SQAs) pose practical threats to deep neural networks by crafting adversarial perturbations within dozens of queries, only using the model's output scores.  ...  In this way, (1) SQAs are prevented regardless of the model's worst-case robustness; (2) the original model predictions are hardly changed, i.e., no degradation on clean accuracy; (3) the calibration of  ...  As a defense in real-world applications, AAA greatly mitigates the adversarial threat without requiring huge computational burden.  ... 
arXiv:2205.12134v1 fatcat:4b3tqbamn5e4pk7n5ppnimyo2i

DeepGauge: multi-granularity testing criteria for deep learning systems

Lei Ma, Yang Liu, Jianjun Zhao, Yadong Wang, Felix Juefei-Xu, Fuyuan Zhang, Jiyuan Sun, Minhui Xue, Bo Li, Chunyang Chen, Ting Su, Li Li
2018 Proceedings of the 33rd ACM/IEEE International Conference on Automated Software Engineering - ASE 2018  
However, a plethora of studies have shown that the state-of-the-art DL systems suffer from various vulnerabilities which can lead to severe consequences when applied to real-world applications.  ...  potentially hinder its real-world deployment.  ...  We gratefully acknowledge the support of NVIDIA AI Tech Center (NVAITC) to our research. We also appreciate the anonymous reviewers for their insightful and constructive comments.  ... 
doi:10.1145/3238147.3238202 dblp:conf/kbse/MaJZSXLCSLLZW18 fatcat:uvf2ugxmnnhblk5lhumzlagpu4

Understanding Local Robustness of Deep Neural Networks under Natural Variations [chapter]

Ziyuan Zhong, Yuchi Tian, Baishakhi Ray
2021 Lecture Notes in Computer Science  
) tool to automatically identify the non-robust points.  ...  This work aims to bridge this gap.To this end, we study the local per-input robustness properties of the DNNs and leverage those properties to build a white-box (DeepRobust-W) and a black-box (DeepRobust-B  ...  Usage Scenario DeepRobust-W/B works in a real-world setting where a customer/user runs a pre-trained DNN model in real-time which constantly receives inputs and wants to test if the prediction of the DNN  ... 
doi:10.1007/978-3-030-71500-7_16 fatcat:lp6tjzbjkvbhzhw2d45qyetyry

The RFML Ecosystem: A Look at the Unique Challenges of Applying Deep Learning to Radio Frequency Applications [article]

Lauren J. Wong, William H. Clark IV, Bryse Flowers, R. Michael Buehrer, Alan J. Michaels, William C. Headley
2020 arXiv   pre-print
deep machine learning systems in real-world wireless communication applications.  ...  A major driver for the usage of deep machine learning in the context of wireless communications is that little, to no, a priori knowledge of the intended spectral environment is required, given that there  ...  Finally, while not as optimal as real world data, augmented datasets aim to provide a "best of both worlds" approach by minimizing the limitations of synthetic datasets (i.e. real-world model accuracy)  ... 
arXiv:2010.00432v1 fatcat:mxnvorh5wrfwzmxg4ezpbj4xve

Is Approximation Universally Defensive Against Adversarial Attacks in Deep Neural Networks? [article]

Ayesha Siddique, Khaza Anuarul Hoque
2021 arXiv   pre-print
Towards this, we present an extensive adversarial robustness analysis of different approximate DNN accelerators (AxDNNs) using the state-of-the-art approximate multipliers.  ...  Very recently, the inexact nature of approximate components, such as approximate multipliers have also been reported successful in defending adversarial attacks on DNNs models.  ...  THREAT MODEL In this section, a threat model is presented for exploring the adversarial robustness of AxDNNs. A.  ... 
arXiv:2112.01555v1 fatcat:eopeppoqaffm3ldhxwqhb3gelm

PRoA: A Probabilistic Robustness Assessment against Functional Perturbations [article]

Tianle Zhang, Wenjie Ruan, Jonathan E. Fieldsend
2022 arXiv   pre-print
and frequently occurring in the real world.  ...  However, existing robustness verification methods are not sufficiently practical for deploying machine learning systems in the real world.  ...  systems in the real world.  ... 
arXiv:2207.02036v1 fatcat:frl7wskdmjhw5inedib46w33d4

Universal Adversarial Perturbations for Speech Recognition Systems

Paarth Neekhara, Shehzeen Hussain, Prakhar Pandey, Shlomo Dubnov, Julian McAuley, Farinaz Koushanfar
2019 Interspeech 2019  
The existence of such perturbations poses a threat to machine learning models in real world settings since the adversary may simply add the same pre-computed universal perturbation to a new image and cause  ...  The existence of universal adversarial perturbations (described below) can pose a more serious threat to ASR systems in real-world settings since the adversary may simply add the same pre-computed universal  ... 
doi:10.21437/interspeech.2019-1353 dblp:conf/interspeech/NeekharaHPDMK19 fatcat:wzfjggj3j5bsxgbxql7zkchlyi
« Previous Showing results 1 — 15 out of 763 results