Adversarial training has proved to be competitive against supervised learning methods on computer vision tasks. However, studies have mainly been confined to generative tasks such as image synthesis. In this paper, we apply adversarial training techniques to the discriminative task of learning a steganographic algorithm. Steganography is a collection of techniques for concealing the existence of information by embedding it within a non-secret medium, such as cover texts or images. We show that adversarial training can produce robust steganographic techniques: our unsupervised training scheme produces a steganographic algorithm that competes with state-of-the-art steganographic techniques. We also show that supervised training of our adversarial model produces a robust steganalyzer, which performs the discriminative task of deciding if an image contains secret information. We define a game between three parties, Alice, Bob and Eve, in order to simultaneously train both a steganographic algorithm and a steganalyzer. Alice and Bob attempt to communicate a secret message contained within an image, while Eve eavesdrops on their conversation and attempts to determine if secret information is embedded within the image. We represent Alice, Bob and Eve by neural networks, and validate our scheme on two independent image datasets, showing our novel method of studying steganographic problems is surprisingly competitive against established steganographic techniques.
Jamie Hayes (University College London)
George Danezis (University College London)
George Danezis is a Professor of Security and Privacy Engineering at the Department of Computer Science of University College London, and Head of the Information Security Research Group. He has been working on anonymous communications, privacy enhancing technologies (PET), and traffic analysis since 2000. He has previously been a researcher for Microsoft Research, Cambridge; a visiting fellow at K.U.Leuven (Belgium); and a research associate at the University of Cambridge (UK), where he also completed his doctoral dissertation under the supervision of Prof. R.J. Anderson. His theoretical contributions to the Privacy Technologies field include the established information theoretic and other probabilistic metrics for anonymity and pioneering the study of statistical attacks against anonymity systems. On the practical side he is one of the lead designers of the anonymous mail system Mixminion, as well as Minx, Sphinx, Drac and Hornet; he has worked on the traffic analysis of deployed protocols such as Tor. His current research interests focus around secure communications, high-integirty systems to support privacy, smart grid privacy, peer-to-peer and social network security, as well as the application of machine learning techniques to security problems. He has published over 70 peer-reviewed scientific papers on these topics in international conferences and journals. He was the co-program chair of ACM Computer and Communications Security Conference in 2011 and 2012, IFCA Financial Cryptography and Data Security in 2011, the Privacy Enhancing Technologies Workshop in 2005 and 2006. He sits on the PET Symposium board and ACM CCS Steering committee and he regularly serves in program committees of leading conferences in the field of privacy and security. He is a fellow of the British Computing Society since 2014.
More from the Same Authors
2018 Poster: Contamination Attacks and Mitigation in Multi-Party Machine Learning »
Jamie Hayes · Olga Ohrimenko