Filters








9 Hits in 9.2 sec

Protecting Against Image Translation Deepfakes by Leaking Universal Perturbations from Black-Box Neural Networks [article]

Nataniel Ruiz, Sarah Adel Bargal, Stan Sclaroff
2020 arXiv   pre-print
In this work, we develop efficient disruptions of black-box image translation deepfake generation systems.  ...  We are the first to demonstrate black-box deepfake generation disruption by presenting image translation formulations of attacks initially proposed for classification models.  ...  Instead of detecting deepfakes after the fact, Ruiz et al. [32] recently proposed using white-box adversarial attacks to protect an image from modification by image translation networks.  ... 
arXiv:2006.06493v1 fatcat:dpcglkyqozh5rkxkjypfd6iaou

Restricted Black-box Adversarial Attack Against DeepFake Face Swapping [article]

Junhao Dong, Yuan Wang, Jianhuang Lai, Xiaohua Xie
2022 arXiv   pre-print
Moreover, we demonstrate that the proposed algorithm can be generalized to offer face image protection against various face translation methods.  ...  Our method is built on a substitute model persuing for face reconstruction and then transfers adversarial examples from the substitute model directly to inaccessible black-box DeepFake models.  ...  image translation models in the restricted black-box scenario.  ... 
arXiv:2204.12347v1 fatcat:76rcbkxadfbmpl2j7a3pekl3gi

Generative Models for Security: Attacks, Defenses, and Opportunities [article]

Luke A. Bauer, Vincent Bindschaedler
2021 arXiv   pre-print
Generative models learn the distribution of data from a sample dataset and can then generate new data instances.  ...  Finally, we discuss new threats due to generative models: the creation of synthetic media such as deepfakes that can be used for disinformation.  ...  Also, within all of these categories, there are different attack scenarios and threat models such as whether the attacker has black-box or white-box access to the model.  ... 
arXiv:2107.10139v2 fatcat:wjb4dcdpvveztd2h4aretus56a

A Survey on Adversarial Attack in the Age of Artificial Intelligence

Zixiao Kong, Jingfeng Xue, Yong Wang, Lu Huang, Zequn Niu, Feng Li, Weizhi Meng
2021 Wireless Communications and Mobile Computing  
Facing the increasingly complex neural network model, this paper focuses on the fields of image, text, and malicious code and focuses on the adversarial attack classifications and methods of these three  ...  Image l 0 One-pixel 2017 [38] Black box Targeted Image l 0 C&W 2017 [39] White box Nontargeted Image l 0, l 2, l ∞ Universal perturbations 2017 [40] White box Nontargeted Universal  ...  DISTFLIP 2019 [31] Black box Nontargeted Text Gradient UPSET 2011 [32] Black box Targeted Universal l ∞ L-BFGS 2014 [33] White and black box Targeted Image l ∞ FGSM-based 2015 [34] White  ... 
doi:10.1155/2021/4907754 fatcat:rm6xcf6ryrh6ngro4sl5ifprgy

Trustworthy AI: From Principles to Practices [article]

Bo Li, Peng Qi, Bo Liu, Shuai Di, Jingen Liu, Jiquan Pei, Jinfeng Yi, Bowen Zhou
2022 arXiv   pre-print
However, many current AI systems are found vulnerable to imperceptible attacks, biased against underrepresented groups, lacking in user privacy protection.  ...  To unify currently available but fragmented approaches toward trustworthy AI, we organize them in a systematic approach that considers the entire lifecycle of AI systems, ranging from data acquisition  ...  It is shown to further provide defense against black-box attacks which do not require knowledge of the model parameters This can help defend against black-box attacks, which do not require knowledge of  ... 
arXiv:2110.01167v2 fatcat:2u7hqdrfujc5lbcsmwpmxxsd74

The State of AI Ethics Report (June 2020) [article]

Abhishek Gupta
2020 arXiv   pre-print
Artificial intelligence has become the byword for technological progress and is being used in everything from helping us combat the COVID-19 pandemic to nudging our attention in different directions as  ...  Our staff has worked tirelessly over the past quarter surfacing signal from the noise so that you are equipped with the right tools and knowledge to confidently tread this complex yet consequential domain  ...  They are encoded deep within our biological systems and suffer from the same lack of explainability as decisions made by artificial neural networks, the so-called black box problem.  ... 
arXiv:2006.14662v1 fatcat:q76dnqzh4ja5pofurjmpmyeyey

On the Opportunities and Risks of Foundation Models [article]

Rishi Bommasani, Drew A. Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, Michael S. Bernstein, Jeannette Bohg, Antoine Bosselut, Emma Brunskill, Erik Brynjolfsson, Shyamal Buch (+102 others)
2022 arXiv   pre-print
Homogenization provides powerful leverage but demands caution, as the defects of the foundation model are inherited by all the adapted models downstream.  ...  This report provides a thorough account of the opportunities and risks of foundation models, ranging from their capabilities (e.g., language, vision, robotics, reasoning, human interaction) and technical  ...  Foundation Models (CRFM), a center at Stanford University borne out of the Stanford Institute for Human-Centered Artificial Intelligence (HAI).  ... 
arXiv:2108.07258v3 fatcat:kohwrwk2ybf7fd7wsuz2gp65ki

D1.3 Cyberthreats and countermeasures

Mark Ryan, Kevin Macnish, Tally Hatzakis
2019
In some areas, artificial intelligence has become powerful to the point that trained models have been withheld from the public over concerns of potential malicious use.  ...  This black box inference technique works very well against models generated by online machinelearning-as-a-service offerings, such as those available from Google and Amazon.  ...  Dropout in neural networks.  ... 
doi:10.21253/dmu.7951292.v1 fatcat:w3z55dymsjcwfkhp7opx4pvhui

Machine Learning Based User Modeling for Enterprise Security and Privacy Risk Mitigation

Preetam Kumar Dutta
2019
log files created by the technology itself.  ...  The challenges are diverse, ranging from malicious parties to vulnerable hardware.  ...  ese black box systems can only be tested in situ. BUBA was designed for exactly that purpose.  ... 
doi:10.7916/d8-w2k6-gp40 fatcat:5rhn467gdbghjiv3i76fyh6qtu