Pierre Roy

Today, huge collections of multimedia contents are available over the Internet, on home computers or even on portable devices. Databases of several hundred thousand audio and/or video files have become commonplaces. This context raises many issues regarding data organization, user queries and, more generally, access to those contents. One of the most challenging issues we face today is to design algorithms and tools for providing users with new efficient ways of accessing their own or other people's contents. These tools have to scale-up to very large collections, and be robust enough to integrate seamlessly in home or portable devices.

Addressing this problem requires to tackle the whole chain from high-level specification down to efficient implementation on dedicated platforms. To handle the inherent combinatorial complexity of these problems, I design novel Operation Research and Artificial Intelligence exploration techniques, borrowing from the fields of intelligent search, pruning strategies and constraint satisfaction. Not only do the algorithms have to be efficient, but they are also designed to be easily distributed and vectorized to take advantage of the power of networks and of modern hardware features. Finally, I also study the efficient implementation of these techniques with an emphasis on code verification, robustness and scale-up.