|Abstract:Security-oriented applications of signal processing have received increasing attention in the last years. Digital watermarking, steganography and steganalysis, multimedia forensics, biometric signal processing, video-surveillance, are just a few examples of such an interest. In many cases, though, researchers (especially signal processing researchers) have failed to recognize the single most unique feature behind any security-oriented application, i.e. the presence of one or more adversaries aiming at making the system fail. One of the most evident consequences is that security requirements are misunderstood. This has long been the case, for instance, in digital watermarking, where it took several years to recognize that robustness and security are contrasting requirements calling for the adoption of different countermeasures. In a similar way, security issues in biometric research are often neglected, privileging pattern recognition issues more related to robustness than security. Similar concerns apply to multimedia forensics, network flow analysis, spam filtering.|
Even when the need to cope with the actions of a malevolent adversary is taken into account, the proposed solutions are often ad-hoc, failing to provide a unifying view of the challenges that such a scenario poses from a signal processing perspective. Times are ripe to go beyond this limited view and lay the basis for a general theory that takes into account the impact that the presence of an adversary has on the design of effective signal processing tools, i.e. a theory of adversarial signal processing.
It is the goal of this talk to present the challenges and opportunities that the development of such a theory poses and to summarize some scattered steps made in this direction in various fields including watermarking, multimedia forensics, biometry, adversary aware classification.
|Abstract: Network privacy has for a long time been conflated with encrypting one's traffic. But currently deployed encryption technologies cannot hide everything. Who is talking to whom, the timing of encrypted traffic and the source and destination of connections may provide invaluable information to observers through traffic analysis. In this talk I discuss anonymity protocols that were designed to prevent traffic analysis. Over the past 10 years they have gone from being research prototypes, to being widely deployed. In parallel, advances in traffic analysis through the application of modern statistical techniques, have demonstrated there are fundamental limits to anonymity. Yet, following deployment many entities and governments have found it easier to block anonymous communications rather tracing them, leading to a renewed interest in covert communications. The developments in this field will be shaping our understanding of information hiding as a research field, but also have deep policy and political ramifications.|
|Abstract: The application of electronic technology to the problem of recognizing people by their voices dates to World War II. Technical developments over the last 70 years have lead to the availability of fully automated systems for recognizing speakers in many applications, but recognition within a forensic context (courts of law) presents a variety of challenges beyond the purely technical. This talk will consider the technical, scientific, legal and philosophical issues, controversies and advancements currently impacting the forensic speaker recognition community. Because of the presenter's background, the talk will be biased towards applications in the U.S., but leading contributions from European, Asian and Pacific countries will also be discussed.|