Filters








9,962 Hits in 2.8 sec

Subnet Replacement: Deployment-stage backdoor attack against deep neural networks in gray-box setting [article]

Xiangyu Qi, Jifeng Zhu, Chulin Xie, Yong Yang
2021 arXiv   pre-print
of the victim model is available but the adversaries do not have any knowledge of parameter values.  ...  Considering the realistic practicability, we abandon the strong white-box assumption widely adopted in existing studies, instead, our algorithm works in a gray-box setting, where architecture information  ...  original capacity, we achieve over 95% attack success rate (over 95% of test samples successfully elicit the adversarial model behavior in the presence of backdoor trigger) with less than 1% loss of clean  ... 
arXiv:2107.07240v1 fatcat:dyrwb62b4jfalnzp6lm6xefxci

Counterterrorism for Cyber-Physical Spaces: A Computer Vision Approach

Cascavilla Giuseppe, Johann Slabber, Fabio Palomba, Dario Di Nucci, Damian A. Tamburri, Willem-Jan Van Den Heuvel
2020 Zenodo  
This paper addresses the aforementioned issue with ALTer a framework featuring computer vision and Generative Adversarial Neural Networks (GANs) over terrorist scenarios.  ...  Moreover, the usage of our synthetic scenarios elicited from GTAV is promising in building datasets for cybersecurity and Cyber-Threat Intelligence (CTI) featuring simulated video gaming platforms.  ...  Table 1 1 lists the parameters we elicited from the state of the art grouped in categories namely: (i) the Environment where the terrorist attack happen, (ii) the Event happening in the environment, and  ... 
doi:10.5281/zenodo.4534150 fatcat:shq6icuabvf4xgegp2lm52lsqy

Counterterrorism for Cyber-Physical Spaces: A Computer Vision Approach

Cascavilla Giuseppe, Johann Slabber, Fabio Palomba, Dario Di Nucci, Damian A. Tamburri, Willem-Jan Van Den Heuvel
2020 Zenodo  
This paper addresses the aforementioned issue with ALTer a framework featuring computer vision and Generative Adversarial Neural Networks (GANs) over terrorist scenarios.  ...  Moreover, the usage of our synthetic scenarios elicited from GTAV is promising in building datasets for cybersecurity and Cyber-Threat Intelligence (CTI) featuring simulated video gaming platforms.  ...  Table 1 1 lists the parameters we elicited from the state of the art grouped in categories namely: (i) the Environment where the terrorist attack happen, (ii) the Event happening in the environment, and  ... 
doi:10.5281/zenodo.4534149 fatcat:luhzeaydvfempc5foaqxhhsyay

Simulation-Based Cyber Data Collection Efficacy [article]

David Thaw, Bret Barkley, Gerry Bella, Carrie Gardner
2019 arXiv   pre-print
Unlike traditional honeypots or honeynets, our experiment utilizes a full-scale operational network to model a small business environment.  ...  Given network activity appropriate for its context, results support the conclusion that no actors where able to break in, despite only default security settings.  ...  Our data also shows evidence of potential adversaries gathering information about our network, through port scanning.  ... 
arXiv:1905.09336v1 fatcat:3rc3hmnqszdvbd74zwtax3v36y

Discussion of "Network routing in a dynamic environment"

Andrew C. Thomas, Stephen E. Fienberg
2011 Annals of Applied Statistics  
Discussion of "Network routing in a dynamic environment" by N.D. Singpurwalla [arXiv:1107.4852]  ...  As we mentioned, social network analysts typically use measures of "centrality" on a graph to elicit information about the role of a node or an edge on the network, often deriving these from the role of  ...  Rather than a decision-theoretic treatment, we consider a method based in part on social network analytical methods, namely, that the deployment pattern of IEDs induces a subgraph on a full road network  ... 
doi:10.1214/11-aoas453a fatcat:fdystcb56ffzteq2p7tsjlh54y

A Gaming Environment for Resilient Network Design and Adversarial Co-Evolution Modeling

YMarco Carvalho, Adrian Granados, James McLane, Evan Stoner
2014 Lecture Notes on Information Theory  
This paper describes the implementation of an adversarial gaming environment for the development, test, and evaluation of Resilience Network and Electronic Warfare procedures.  ...  Recognizing the co-evolutionary nature of the attacker, we argue that, especially in the case of adversarial environments, an intelligent and adaptive attacker model must be considered for the test and  ...  Each event elicits the appropriate response from the system to emulate network and link conditions based on EMANE's physical and propagation models.  ... 
doi:10.12720/lnit.2.1.92-97 fatcat:2jcaq743hjc6lf6a3qykbj3oae

Design of intentional backdoors in sequential models [article]

Zhaoyuan Yang, Naresh Iyer, Johan Reimann, Nurali Virani
2019 arXiv   pre-print
Challenges with network size and unintentional triggers are identified and analogies with adversarial examples are also discussed.  ...  In contrast to adversarial examples, backdoor or trojan attacks embed surgically modified samples with targeted labels in the model training process to cause the targeted model to learn to misclassify  ...  In these attacks, the adversary designs appropriate triggers that can be used to elicit unexpected and unanticipated behavior from a seemingly honest model.  ... 
arXiv:1902.09972v1 fatcat:f44cjjodmra3be5umshlczzkxa

Deep Architecture Enhancing Robustness to Noise, Adversarial Attacks, and Cross-corpus Setting for Speech Emotion Recognition [article]

Siddique Latif, Rajib Rana, Sara Khalifa, Raja Jurdak, Björn W. Schuller
2020 arXiv   pre-print
In this paper we propose a deeper neural network architecture wherein we fuse DenseNet, LSTM and Highway Network to learn powerful discriminative features which are robust to noise.  ...  We comprehensively evaluate the architecture coupled with data augmentation against (1) noise, (2) adversarial attacks and (3) cross-corpus settings.  ...  Adversarial attacks are developed by malicious adversaries to cra adversarial examples by the addition of unperceived perturbation to elicit wrong responses from machine learning (ML) models.  ... 
arXiv:2005.08453v3 fatcat:bfjesm7wz5gujn7pye67fu5fyi

Deep Architecture Enhancing Robustness to Noise, Adversarial Attacks, and Cross-Corpus Setting for Speech Emotion Recognition

Siddique Latif, Rajib Rana, Sara Khalifa, Raja Jurdak, Björn W. Schuller
2020 Interspeech 2020  
In this paper we propose a deeper neural network architecture wherein we fuse Dense Convolutional Network (DenseNet), Long shortterm memory (LSTM) and Highway Network to learn powerful discriminative features  ...  We comprehensively evaluate the architecture coupled with data augmentation against (1) noise, (2) adversarial attacks and (3) cross-corpus settings.  ...  Adversarial attacks are developed by malicious adversaries to craft adversarial examples by the addition of unperceived perturbation to elicit wrong responses from machine learning (ML) models.  ... 
doi:10.21437/interspeech.2020-3190 dblp:conf/interspeech/LatifRKJS20 fatcat:qac6duzm3vh4jffneieq6mvsni

An Overview and Prospective Outlook on Robust Training and Certification of Machine Learning Models [article]

Brendon G. Anderson, Tanmay Gautam, Somayeh Sojoudi
2022 arXiv   pre-print
As learning algorithms become increasingly more popular in data-driven control systems, their robustness to data uncertainty must be ensured in order to maintain reliable safety-critical operations.  ...  In this discussion paper, we survey recent research surrounding robustness of machine learning models.  ...  Here studies have shown that slight manipulations in the input images can elicit misclassifications in neural networks with high confidence (Szegedy et al., 2014; Moosavi-Dezfooli et al., 2016; Goodfellow  ... 
arXiv:2208.07464v2 fatcat:qpyaolc57baxtoxy3nx4qufgfu

Primer – A Tool for Testing Honeypot Measures of Effectiveness [article]

Jason M. Pittman, Kyle Hoffpauir, Nathan Markle
2020 arXiv   pre-print
The measures quantify a dynamic honeypot's effectiveness in fingerprinting its environment, capturing valid data from adversaries, deceiving adversaries, and intelligently monitoring itself and its surroundings  ...  We outline the design of the tool and provide results in the form of quantitative calibration data.  ...  Thus, there is no guarantee of an attack to elicit interaction and utilization in a honeypot.  ... 
arXiv:2011.00582v1 fatcat:22svs7z2tbfxjhi3oeo6r5lkri

Gradient-Based Adversarial and Out-of-Distribution Detection [article]

Jinsol Lee, Mohit Prabhushankar, Ghassan AlRegib
2022 arXiv   pre-print
We introduce confounding labels -- labels that differ from normal labels seen during training -- in gradient generation to probe the effective expressivity of neural networks.  ...  state-of-the-art methods for adversarial and out-of-distribution detection.  ...  The separation in the ranges of gradient magnitudes, shown in Fig. 3a , between original images of CIFAR-10 test set and their adversarial versions is more evident in some parts of the network than others  ... 
arXiv:2206.08255v2 fatcat:os55tbr46zaitm2sen4pqmcz24

Multi-task Maximum Entropy Inverse Reinforcement Learning [article]

Adam Gleave, Oliver Habryka
2018 arXiv   pre-print
Experiments show our approach can perform one-shot imitation learning in a gridworld environment that single-task IRL algorithms need hundreds of demonstrations to solve.  ...  Prior work, built on Bayesian IRL, is unable to scale to complex environments due to computational constraints.  ...  We conjecture this limitation in adversarial IRL is related to the well-known problem of mode collapse in generative adversarial networks (GAN).  ... 
arXiv:1805.08882v2 fatcat:tzap2hkn3jfazcxj2gq2hffnaa

DOS Attacks on TCP/IP Layers in WSN

2013 International journal of computer networks and communications security  
While the set of challenges in sensor networks are diverse, we focus on security of Wireless Sensor Network in this paper. First, we propose some of the security goal for Wireless Sensor Network.  ...  In contrast to this crucial objective of sensor network management, a Denial of Service (DoS) attack targets to degrade the efficient use of network resources and disrupts the essential services in the  ...  For the proper functioning of WSN, especially in malicious environments, security mechanisms become essential for all kinds of sensor networks.  ... 
doi:10.47277/ijcncs/1(2)1 fatcat:wnmt4tcymzfnxpwj7zx2zgfehe

Interpreting Multimodal Machine Learning Models Trained for Emotion Recognition to Address Robustness and Privacy Concerns

Mimansa Jaiswal
2020 PROCEEDINGS OF THE THIRTIETH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE AND THE TWENTY-EIGHTH INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE CONFERENCE  
Yet, these data contain sensitive information that could be used by mobile applications without user's consent or, maliciously, by an eavesdropping adversary.  ...  These predicted emotions are used in variety of downstream applications: (a) generating more human like dialogues, (b) predicting mental health issues, and (c) hate speech detection and intervention.  ...  To this end, we use adversarial networks to decorrelate stress modulations from emotion representations.  ... 
doi:10.1609/aaai.v34i10.7130 fatcat:l24ba4cgbjgzzkmkzjceosrjyu
« Previous Showing results 1 — 15 out of 9,962 results