Information hiding refers to a subfield at the intersection of systems security, signal processing, and information theory. It unites research on methods to communicate over channels that are invisible to an observer, and often were not intended as communication channels.

Since the early 1990s, the information hiding community has generated a large body of knowledge, from theoretical models for covert information transmission down to practical realizations of information hiding methods for various embedding domains, witnessed by several thousand publications. Two key applications were the main drivers behind this work: Gus Simon's prisoners' problem for steganography and steganalysis, and Digital Rights Management (DRM) for content watermarking.

Over the years, these two applications became less attractive, but new applications have emerged that require similar technology: covert channels can now be used to leak secrets out of heavily guarded servers; to resist traffic analysis and support censorship resistance; to transfer watermarked information between a television set and a smartphone for targeted advertising, posing a risk to privacy; to infer the nature of processing performed on servers or smartphones; or can even be related to adversarial examples against machine learning-based vision or speech recognition systems. All these applications critically depend on methods of information hiding, yet they have largely been developed without direct interaction with the information hiding community.

This Dagstuhl Seminar aims at bridging this gap; it brings together researchers from the traditional information hiding discipline with researchers working on novel problems and applications. The seminar will address three specific topics. First, it will explore the interplay between information hiding and privacy: on the one hand, novel hiding methods may invade privacy, while on the other hand privacy enhancing technologies can be built using methods from information hiding. Second, the seminar will investigate covert channels in novel architectures and their implications on system security. Third, the seminar will study the interplay between information hiding and the vulnerabilities of machine learning to adversarial inputs.

Creative Commons BY 3.0 DE

Rainer Böhme, Stefan Katzenbeisser, Jakub Szefer, and Carmela Troncoso