ERC FUNDED PROJECTS

One of the primary and most appealing goals of computer vision is to automatically understand the content of images on a cognitive level. Ultimately we want to have computers interpret images as we humans do, recognizing all the objects, scenes, and people as well as their relations as they appear in natural images or video. With this project, I want to advance the state of the art in this field in two directions, which I believe to be crucial to build the next generation of image understanding tools. First, novel more robust yet descriptive image representations will be designed, that incorporate the intrinsic structure of images. These should already go a long way towards removing irrelevant sources of variability while capturing the essence of the image content. I believe the importance of further research into image representations is currently underestimated within the research community, yet I claim this is a crucial step with lots of opportunities good learning cannot easily make up for bad features. Second, weakly supervised methods to learn from multimodal input (especially the combination of images and text) will be investigated, making it possible to leverage the large amount of weak annotations available via the internet. This is essential if we want to scale the methods to a larger number of object categories (several hundreds instead of a few tens). As more data can be used for training, such weakly supervised methods might in the end even come on par with or outperform supervised schemes. Here we will call upon the latest results in semi-supervised learning, datamining, and computational linguistics.

1 538 380 €

Start date: 2010-02-01, End date: 2015-01-31

Side-channel attacks are an important threat against cryptographic implementations in which an adversary takes advantage of physical leakages, such as the power consumption of a smart card, in order to recover secret information. By circumventing the models in which standard security proofs are obtained, they can lead to powerful attacks against a large class of devices. As a consequence, formalizing implementation security and efficiently preventing side-channel attacks is one of the most challenging open problems in modern cryptography. Physical attacks imply new optimization criteria, with potential impact on the way we conceive algorithms and the way we design circuits. By putting together mathematical and electrical engineering problems, just as they are raised in reality, the CRASH project is expected to develop concrete basements for the next generation of cryptographic algorithms and their implementation. For this purpose, three main directions will be considered. First, we will investigate sound evaluation tools for side-channel attacks and validate them on different prototype chips. Second, we will consider the impact of physical attacks on the mathematical aspects of cryptography, both destructively (i.e. by developing new attacks and advanced cryptanalysis tools) and constructively (i.e. by investigating new cipher designs and security proof techniques). Third, we will evaluate the possibility to integrate physical security analysis into the design tools of integrated circuits (e.g. in order to obtain “physical security aware” compilers). Summarizing, this project aims to break the barrier between the abstractions of mathematical cryptography and the concrete peculiarities of physical security in present microelectronic devices. By considering the system and algorithmic issues in a unified way, it is expected to get rid of the incompatibilities between the separate formalisms that are usually considered in order to explain these concurrent realities.

1 498 874 €

Start date: 2011-10-01, End date: 2016-09-30

The scope of this project is the automatic design of robot swarms. Swarm robotics is an appealing approach to the coordination of large groups of robots. Up to now, robot swarms have been designed via some labor-intensive process. My goal is to advance the state of the art in swarm robotics by developing the DEMIURGE: an intelligent system that is able to design and realize robot swarms in a totally integrated and automatic way The DEMIURGE is a novel concept. Starting from requirements expressed in a specification language that I will define, the DEMIURGE will design all aspects of a robot swarm - hardware and control software. The DEMIURGE will cast a design problem into an optimization problem and will tackle it in a computation-intensive way. In this project, I will study different control software structures, optimization algorithms, ways to specify requirements, validation protocols, on-line adaptation mechanisms and techniques for re-design at run time.

2 000 000 €

Start date: 2016-10-01, End date: 2021-09-30

Contemporary microprocessors seek at improving performance through thread-level parallelism by co-executing multiple threads on a single microprocessor chip. Projections suggest that future processors will feature multiple tens to hundreds of threads, hence called many-thread processors. Many-thread processors, however, lead to non-dependable performance: co-executing threads affect each other s performance in unpredictable ways because of resource sharing across threads. Failure to deliver dependable performance leads to missed deadlines, priority inversion, unbalanced parallel execution, etc., which will severely impact the usage model and the performance growth path for many important future and emerging application domains (e.g., media, medical, datacenter). DPMP envisions that performance introspection using a cycle accounting architecture that tracks per-thread performance, will be the breakthrough to delivering dependable performance in future many-thread processors. To this end, DPMP will develop a hardware cycle accounting architecture that estimates single-thread progress during many-thread execution. The ability to track per-thread progress enables system software to deliver dependable performance by assigning hardware resources to threads depending on their relative progress. Through this cooperative hardware-software approach, this project addresses a fundamental problem in multi-threaded ad multi/many-core processing.

1 389 000 €

Start date: 2010-10-01, End date: 2016-09-30

"Linear programming has proved to be an invaluable tool both in theory and practice. Semidefinite programming surpasses linear programming in terms of expressivity while remaining tractable. This project proposal investigates the modeling power of linear and semidefinite programming, in the context of combinatorial optimization. Within the emerging framework of extended formulations (EFs), I seek a decisive answer to the following question: Which problems can be modeled by a linear or semidefinite program, when the number of constraints and variables are limited? EFs are based on the idea that one should choose the ""right"" variables to model a problem. By extending the set of variables of a problem by a few carefully chosen variables, the number of constraints can in some cases dramatically decrease, making the problem easier to solve. Despite previous high-quality research, the theory of EFs is still on square one. This project proposal aims at (i) transforming our current zero-dimensional state of knowledge to a truly three-dimensional state of knowledge by pushing the boundaries of EFs in three directions (models, types and problems); (ii) using EFs as a lens on complexity by proving strong consequences of important conjectures such as P != NP, and leveraging strong connections to geometry to make progress on the log-rank conjecture. The proposed methodology is: (i) experiment-aided; (ii) interdisciplinary; (iii) constructive."

1 455 479 €

Start date: 2014-09-01, End date: 2019-08-31

The goal of IMPaCT is to turn Multi-Party Computation (MPC) from a stage in which we are beginning to obtain practical feasibility results, to a stage in which we have fully practical systems. It has long been acknowledged that MPC has the potential to provide a transformative change in the way security solutions are enabled. As it presently stands this is currently only possible in limited applications; deployments in restricted scenarios are beginning to emerge. However, in turning MPC into a fully practical technology a number of key scientific challenges need to be solved; many of which have not yet even been considered in the theoretical literature. The IMPaCT project aims to address this scientific gap, bridge it, and so provide the tools for a future road-map in which MPC can be deployed as a widespread tool, as ubiquitous as encryption and digital signatures are today. Our scientific approach will be to investigate new MPC protocols and techniques which take into account practical constraints and issues which would arise in future application scenarios. Our work, despite being scientifically rigorous and driven from deep theoretical insight, will be grounded in practical considerations. All systems and protocols proposed will be prototyped so as to ensure that practical real world issues are taken into account. In addition we will use our extensive industrial linkages to ensure a two way dialogue between potential users and the developers of MPC technology; thus helping to embed future impact of the work in IMPaCT. Our workplan is focused around key scientific challenges which we have identified on the road to fully practical MPC applications. These include the design of methodologies to cope with the asynchronicity of networks, how to realistically measure and model MPC protocols performance, how to utilize low round complexity protocols in practice, how to deal with problems with large input sizes (e.g. streaming data), and many more.

2 499 938 €

Start date: 2016-10-01, End date: 2021-09-30

Finite-state devices like finite automata and monoids on finite words, or extensions to trees and infinite objects, are fundamental tools of logic in computer science. There are tens of models in the literature, ranging from finite automata on finite words to weighted automata on infinite trees. Many existing finite-state models share important similarities, like existence of canonical (minimal) devices, or decidability of emptiness, or a logic-automata connection. The first and primary goal of this project is to systematically investigate these similarities, and create a unified theory of finite-state devices, which: 1. covers the whole spectrum of existing finite-state devices, including settings with diverse inputs (e.g. words and trees, or infinite inputs, or infinite alphabets) and diverse outputs (e.g. Boolean like in the classical automata, or numbers like in weighted automata); and 2. sheds light on the correct notion of finite-state device in settings where there is no universally accepted choice or where finite-state devices have not been considered at all. The theory of finite-state devices is one of those fields of theory where even the more advanced results have natural potential for applications. It is surprising and sad how little of this potential is normally realised, with most existing software using only the most rudimentary theoretical techniques. The second goal of the project is to create two tools which use more advanced aspects of the theory of automata to solve simple problems of wide applicability (i.e. at least tens of thousands of users): 1. a system that automatically grades exercises in automata, which goes beyond simple testing, and forces the students to write proofs 2. a system that uses learning to synthesise text transformations (such a search-and-replace, but also more powerful ones) by using examples

1 768 125 €

Start date: 2016-05-01, End date: 2021-04-30

A critical component of computational processing of data sets is the `preprocessing' or `compression' step which is the computation of a \emph{succinct, sufficiently accurate} representation of the given data. Preprocessing is ubiquitous and a rigorous mathematical understanding of preprocessing algorithms is crucial in order to reason about and understand the limits of preprocessing. Unfortunately, there is no mathematical framework to analyze and objectively compare two preprocessing routines while simultaneously taking into account `all three dimensions' -- -- the efficiency of computing the succinct representation, -- the space required to store this representation, and -- the accuracy with which the original data is captured in the succinct representation. ``The overarching goal of this proposal is the development of a mathematical framework for the rigorous analysis of preprocessing algorithms. '' We will achieve the goal by designing new algorithmic techniques for preprocessing, developing a framework of analysis to make qualitative comparisons between various preprocessing routines based on the criteria above and by developing lower bound tools required to understand the limitations of preprocessing for concrete problems. This project will lift our understanding of algorithmic preprocessing to new heights and lead to a groundbreaking shift in the set of basic research questions attached to the study of preprocessing for specific problems. It will significantly advance the analysis of preprocessing and yield substantial technology transfer between adjacent subfields of computer science such as dynamic algorithms, streaming algorithms, property testing and graph theory.

2 000 000 €

Start date: 2019-05-01, End date: 2024-04-30

The performance and physical attributes of a material and product can be tailored to so far unmatched material strengths and properties by creating new nano fibrous structures from polymers by electrospinning. The electrospinning process uses an electric field to produce charged jets of polymer solutions or melts. Bending instabilities of the jet, caused by the surface charge, lead to extremely high local extension rates of the jet and produce fibres with diameters of the order of a few nanometer that consist of highly aligned polymer strands. However, the biggest unsolved problem of the electrospinning process is the sensitive equilibrium between surface tension, viscosity, elasticity and conductivity of the polymer solutions. These are controlled by molecular parameters as the molar mass, chemical microstructure, conformation in solution or supramolecular structures via intermolecular interactions. The optimal combination of these parameters is, as yet, unknown. Within this project, a novel and unique technical platform will be developed and installed, that is generally capable to image and analyse high speed free surface flows in miniaturised dimensions. This platform will then be utilized to analyse electrospinning process parameters and to connect them to the material properties and the molecular structure of the polymer solution. Only such a fundamental understanding of the relation of these properties to the flow and mass transfer phenomena on the micro-time and -dimensional scale will allow to design in the second part of this project the required structural and material properties of nano-scale fibres for: -novel fibre/matrix composites for the creation of ultra-high-strength hydrogel membranes; -short fibre morphologies created by a novel controlled disruptive spinning process at the boundaries of the parameter space; -tailoring of fibre properties from renewable resources by modification of the chemical side-chain structure of polysaccharides.

1 228 736 €

Start date: 2008-07-01, End date: 2013-06-30