ERC FUNDED PROJECTS

"During the past twenty years, we have witnessed profound technological changes, summarised under the terms of digital revolution or entering the information age. It is evident that these technological changes will have a deep societal impact, and questions of privacy and security are primordial to ensure the survival of a free and open society. Cryptology is a main building block of any security solution, and at the heart of projects such as electronic identity and health cards, access control, digital content distribution or electronic voting, to mention only a few important applications. During the past decades, public-key cryptology has established itself as a research topic in computer science; tools of theoretical computer science are employed to “prove” the security of cryptographic primitives such as encryption or digital signatures and of more complex protocols. It is often forgotten, however, that all practically relevant public-key cryptosystems are rooted in pure mathematics, in particular, number theory and arithmetic geometry. In fact, the socalled security “proofs” are all conditional to the algorithmic untractability of certain number theoretic problems, such as factorisation of large integers or discrete logarithms in algebraic curves. Unfortunately, there is a large cultural gap between computer scientists using a black-box security reduction to a supposedly hard problem in algorithmic number theory and number theorists, who are often interested in solving small and easy instances of the same problem. The theoretical grounds on which current algorithmic number theory operates are actually rather shaky, and cryptologists are generally unaware of this fact. The central goal of ANTICS is to rebuild algorithmic number theory on the firm grounds of theoretical computer science."

1 453 507 €

Start date: 2012-01-01, End date: 2016-12-31

Despite our great expressive skills, we humans lack an easy way of communicating the 3D shapes we imagine, and even more so when it comes to dynamic shapes. Over centuries humans used drawing and sculpture to convey shapes. These tools require significant expertise and time investment, especially if one aims to describe complex or dynamic shapes. With the advent of virtual environments one would expect digital modeling to replace these traditional tools. Unfortunately, conventional techniques in the area have failed, since even trained computer artists still create with traditional media and only use the computer to reproduce already designed content. Could digital media be turned into a tool, even more expressive and simpler to use than a pen, to convey and refine both static and dynamic 3D shapes? This is the goal of this project. Achieving it will make shape design directly possible in virtual form, from early drafting to progressive refinement and finalization of an idea. To this end, models for shape and motion need to be totally rethought from a user-centered perspective . Specifically, we propose the new paradigm of responsive 3D shapes – a novel representation separating morphology from isometric embedding – to define high-level, dynamic 3D content that takes form, is refined, moves and deforms based on user intent, expressed through intuitive interaction gestures. Scientifically, while the problem we address belongs to Computer Graphics, it calls for a new convergence with Geometry, Simulation and Human Computer Interaction. In terms of impact, the resulting “expressive virtual pen” for 3D content will not only serve the needs of artists, but also of scientists and engineers willing to refine their thoughts by interacting with prototypes of their objects of study, educators and media aiming at quickly conveying their ideas, as well as anyone willing to communicate a 3D shape This project thus opens up new horizons for science, technology and society.

2 498 116 €

Start date: 2012-04-01, End date: 2017-03-31

Content creation is a fundamental activity for developing identities in modern individuals. Yet creativity is hardly addressed by computer science. This project addresses the issue of content creation from the perspective of Flow machines. Flow machines are interactive systems that learn how to generate content, text or music, in the user’s style. Thanks to controlled generation mechanisms, the user can then steer the machine to generate content that fits with their intentions. Flow interactions induce a multiplicative effect that boosts creativity and prompts the user to reflect on their own style. This vision stems from the success stories of several computer-assisted musical systems that showed how interactive dialogs with self-learning interactions provoke flow states. To enables full control of stylistic generation, the scientific challenge is the reification of style as a flexible texture. This challenge will be addressed by pursuing three original directions in the fields of statistical learning and combinatorial optimization: 1) the formulation of Markov-based generation as a constraint problem, 2) the development of feature generation techniques for feeding machine learning algorithms and 3) the development of techniques to transform descriptors into controllers. Two large-scale studies will be conducted with well-known creators using these Flow machines, during which the whole creation process will be recorded, stored, and analyzed, providing the first complete chronicles of professional-level artifacts. The artifacts, a music album and a novel, will be published in their respective ecosystems, and the reaction of the audience will be measured and analyzed to further assess the impact of Flow machines on creation. The technologies developed and the pilot studies will serve as pioneering experiments to turn Flow machines into a field of study and explore other domains of creation.

2 240 120 €

Start date: 2012-08-01, End date: 2017-07-31

Every year, software bugs cost hundreds of millions of euros to compagnies and administrations. A number of disasters such as the Ariane 5 first flight failure can are due to faulty softwares. Static analysis aims at computing automatically properties of softwares, so as to prove they are exempt from some class of bugs. In the last ten years, static analysis of numeric intensive applications improved dramatically so that the certification of safety properties like the absence of runtime errors in industrial size control-command, numeric intensive applications, such as Airbus fly-by-wire softwares is now feasible. By contrast, the situation is much worse for memory intensive softwares. Existing static analyzers for such softwares do not scale to large scale softwares, and fail to prove strong invariants on large classes of softwares. These limitations stem from the fact they use a monolithic algebra of logical formulas (or abstract domain). Our proposal is based on the observation that the complex memory properties that need be reasoned about should be decomposed in combinations of simpler properties. Therefore, in static analysis, a powerful memory abstract domain could be designed by combining several simpler domains, specific to common memory usage patterns. The benefit of this novel vision is twofold: first it would make it possible to simplify drastically the design of complex abstract domains required to reason about complex softwares, hereby allowing certification of complex memory intensive softwares by automatic static analysis; second, it would enable to split down and better control the cost of the analyses, thus significantly helping scalability. This shift of focus will bring both theoretical and practical improvements to the program certification field. We propose to build a static analysis framework for reasoning about memory properties, and put it to work on important classes of applications, including large safety critical memory intensive softwares.

1 489 663 €

Start date: 2011-10-01, End date: 2017-09-30