ERC FUNDED PROJECTS

Although we have extensive knowledge of many important processes in cell biology, including information on many of the molecules involved and the physical interactions among them, we still do not understand most of the dynamical features that are the essence of living systems. This is particularly true for the actin cytoskeleton, a major component of the internal architecture of eukaryotic cells. In living cells, actin networks constantly assemble and disassemble filaments while maintaining an apparent stable structure, suggesting a perfect balance between the two processes. Such behaviors are called “dynamic steady states”. They confer upon actin networks a high degree of plasticity allowing them to adapt in response to external changes and enable cells to adjust to their environments. Despite their fundamental importance in the regulation of cell physiology, the basic mechanisms that control the coordinated dynamics of co-existing actin networks are poorly understood. In the AAA project, first, we will characterize the parameters that allow the coupling among co-existing actin networks at steady state. In vitro reconstituted systems will be used to control the actin nucleation patterns, the closed volume of the reaction chamber and the physical interaction of the networks. We hope to unravel the mechanism allowing the global coherence of a dynamic actin cytoskeleton. Second, we will use our unique capacity to perform dynamic micropatterning, to add or remove actin nucleation sites in real time, in order to investigate the ability of dynamic networks to adapt to changes and the role of coupled network dynamics in this emergent property. In this part, in vitro experiments will be complemented by the analysis of actin network remodeling in living cells. In the end, our project will provide a comprehensive understanding of how the adaptive response of the cytoskeleton derives from the complex interplay between its biochemical, structural and mechanical properties.

2 349 898 €

Start date: 2017-09-01, End date: 2022-08-31

The understanding of spatial, social and dynamical organization of large numbers of agents is presently a fundamental issue in modern science. ADORA focuses on problems motivated by biology because, more than anywhere else, access to precise and many data has opened the route to novel and complex biomathematical models. The problems we address are written in terms of nonlinear partial differential equations. The flux-limited Keller-Segel system, the integrate-and-fire Fokker-Planck equation, kinetic equations with internal state, nonlocal parabolic equations and constrained Hamilton-Jacobi equations are among examples of the equations under investigation. The role of mathematics is not only to understand the analytical structure of these new problems, but it is also to explain the qualitative behavior of solutions and to quantify their properties. The challenge arises here because these goals should be achieved through a hierarchy of scales. Indeed, the problems under consideration share the common feature that the large scale behavior cannot be understood precisely without access to a hierarchy of finer scales, down to the individual behavior and sometimes its molecular determinants. Major difficulties arise because the numerous scales present in these equations have to be discovered and singularities appear in the asymptotic process which yields deep compactness obstructions. Our vision is that the complexity inherent to models of biology can be enlightened by mathematical analysis and a classification of the possible asymptotic regimes. However an enormous effort is needed to uncover the equations intimate mathematical structures, and bring them at the level of conceptual understanding they deserve being given the applications motivating these questions which range from medical science or neuroscience to cell biology.

2 192 500 €

Start date: 2017-09-01, End date: 2023-02-28

Recurrent neural networks (RNNs) are general parallel-sequential computers. Some learn their programs or weights. Our supervised Long Short-Term Memory (LSTM) RNNs were the first to win pattern recognition contests, and recently enabled best known results in speech and handwriting recognition, machine translation, etc. They are now available to billions of users through the world's most valuable public companies including Google and Apple. Nevertheless, in lots of real-world tasks RNNs do not yet live up to their full potential. Although universal in theory, in practice they fail to learn important types of algorithms. This ERC project will go far beyond today's best RNNs through novel RNN-like systems that address some of the biggest open RNN problems and hottest RNN research topics: (1) How can RNNs learn to control (through internal spotlights of attention) separate large short-memory structures such as sub-networks with fast weights, to improve performance on many natural short-term memory-intensive tasks which are currently hard to learn by RNNs, such as answering detailed questions on recently observed videos? (2) How can such RNN-like systems metalearn entire learning algorithms that outperform the original learning algorithms? (3) How to achieve efficient transfer learning from one RNN-learned set of problem-solving programs to new RNN programs solving new tasks? In other words, how can one RNN-like system actively learn to exploit algorithmic information contained in the programs running on another? We will test our systems existing benchmarks, and create new, more challenging multi-task benchmarks. This will be supported by a rather cheap, GPU-based mini-brain for implementing large RNNs.

2 500 000 €

Start date: 2017-10-01, End date: 2022-09-30

The project ARTHUS aims at determining the physical origin of Dark Energy: in addition to the energy sources of the standard model of cosmology, effective terms arise through spatially averaging inhomogeneous cosmological models in General Relativity. It has been demonstrated that these additional terms can play the role of Dark Energy on large scales (but they can also mimic Dark Matter on scales of mass accumulations). The underlying rationale is that fluctuations in the Universe generically couple to spatially averaged intrinsic properties of space, such as its averaged scalar curvature, thus changing the global evolution of the effective (spatially averaged) cosmological model. At present, we understand these so- called backreaction effects only qualitatively. The project ARTHUS is directed towards a conclusive quantitative evaluation of these effects by developing generic and non-perturbative relativistic models of structure formation, by statistically measuring the key-variables of the models in observations and in simulation data, and by reinterpreting observational results in light of the new models. It is to be emphasized that there is no doubt about the existence of backreaction effects; the question is whether they are even capable of getting rid of the dark sources (as some models discussed in the literature suggest), or whether their impact is substantially smaller. The project thus addresses an essential issue of current cosmological research: to find pertinent answers concerning the quantitative impact of inhomogeneity effects, a necessary, worldwide recognized step toward high-precision cosmology. If the project objectives are attained, the results will have a far-reaching impact on theoretical and observational cosmology, on the interpretation of astronomical experiments such as Planck and Euclid, as well as on a wide spectrum of particle physics theories and experiments.

2 091 000 €

Start date: 2017-09-01, End date: 2023-08-31