ERC FUNDED PROJECTS

Optimization methods have become an established paradigm to address most Computer Vision challenges including the reconstruction of three-dimensional objects from multiple images, or the tracking of a deformable shape over time. Yet, it has been largely overlooked that optimization approaches are practically useless if they do not come with efficient algorithms to compute minimizers of respective energies. Most existing formulations give rise to non-convex energies. As a consequence, solutions highly depend on the choice of minimization scheme and implementational (initialization, time step sizes, etc.), with little or no guarantees regarding the quality of computed solutions and their robustness to perturbations of the input data. In the proposed research project, we plan to develop optimization methods for Computer Vision which allow to efficiently compute globally optimal solutions. Preliminary results indicate that this will drastically leverage the power of optimization methods and their applicability in a substantially broader context. Specifically we will focus on three lines of research: 1) We will develop convex formulations for a variety of challenges. While convex formulations are currently being developed for low-level problems such as image segmentation, our main effort will focus on carrying convex optimization to higher level problems of image understanding and scene interpretation. 2) We will investigate alternative strategies of global optimization by means of discrete graph theoretic methods. We will characterize advantages and drawbacks of continuous and discrete methods and thereby develop novel algorithms combining the advantages of both approaches. 3) We will go beyond convex formulations, developing relaxation schemes that compute near-optimal solutions for problems that cannot be expressed by convex functionals.

1 985 400 €

Start date: 2010-09-01, End date: 2015-08-31

Swarm intelligence is the discipline that deals with natural and artificial systems composed of many individuals that coordinate using decentralized control and self-organization. In this project, we focus on the design and implementation of artificial swarm intelligence systems for the solution of complex problems. Our current understanding of how to use swarms of artificial agents largely relies on rules of thumb and intuition based on the experience of individual researchers. This is not sufficient for us to design swarm intelligence systems at the level of complexity required by many real-world applications, or to accurately predict the behavior of the systems we design. The goal of the E-SWARM is to develop a rigorous engineering methodology for the design and implementation of artificial swarm intelligence systems. We believe that in the future, swarm intelligence will be an important tool for researchers and engineers interested in solving certain classes of complex problems. To build the foundations of this discipline and to develop an appropriate methodology, we will proceed in parallel both at an abstract level and by tackling a number of challenging problems in selected research domains. The research domains we have chosen are optimization, robotics, networks, and data mining.

2 016 000 €

Start date: 2010-06-01, End date: 2015-05-31

In spite of considerable work in artificial intelligence, machine learning, and pattern recognition in the past 50 years, we have no machine capable of adapting to the physical and social environment with the flexibility, robustness and versatility of a 6-months old human child. Instead of trying to simulate directly the adult s intelligence, EXPLORERS proposes to focus on the developmental principles that give rise to intelligence in infants by re-implementing them in machines. Framed in the developmental/epigenetic robotics research agenda, and grounded in research in developmental psychology, its main target is to build robotic machines capable of autonomously learning and re-using a variety of skills and know-how that were not specified at design time, and with initially limited knowledge of the body and of the environment in which it will operate. This implies several fundamental issues: How can a robot discover its body and its relationships with the physical and social environment? How can it learn new skills without the intervention of an engineer? What internal motivations shall guide its exploration of vast spaces of skills? Can it learn through natural social interactions with humans? How to represent the learnt skills and how can they be re-used? EXPLORERS attacks directly those questions by proposing a series of fundamental scientific and technological advances, including computational intrinsic motivation systems for learning basic sensorimotor skills reused for grounded acquisition of the meaning of new words. This project not only addresses fundamental scientific questions, but also relates to important societal issues: personal home robots are bound to become part of everyday life in the 21st century, in particular as helpful social companions in an aging society. EXPLORERS objectives converge to the challenges implied by this vision: robots will have to be able to adapt and learn new skills in the unknown homes of users who are not engineers.

1 572 215 €

Start date: 2009-12-01, End date: 2015-05-31

The general theme is to explore the connections between reasoning and computations in mathematics. There are two main research directions. The first research direction is a refomulation of Hilbert's program, using ideas from formal, or pointfree topology. We have shown, with multiple examples, that this allows a partial realization of this program in commutative algebra, and a new way to formulate constructive mathematics. The second research direction explores the computational content using type theory and the Curry-Howard correspondence between proofs and programs. Type theory allows us to represent constructive mathematics in a formal way, and provides key insight for the design of proof systems helping in the analysis of the logical structure of mathematical proofs. The interest of this program is well illustrated by the recent work of G. Gonthier on the formalization of the 4 color theorem.

1 912 288 €

Start date: 2010-04-01, End date: 2015-03-31