NETADIS, Background information for applicants SEVENTH FRAMEWORK PROGRAMME THE PEOPLE PROGRAMME Grant agreement for: Initial Training Networks Annex I - “Description of Work” - Extracts (Background information for applicants) Project acronym: NETADIS Project full title: Statistical Physics Approaches to Networks Across Disciplines Project start date: 1 March 2012 Project Abstract Networks are ubiquitous in all aspects of human existence. They manifest themselves in everyday life, they underpin the most advanced information and communication technologies, and they provide a powerful framework to address a wide spectrum of complex problems in the natural sciences, in engineering, and in economics and the social sciences. Networks, network-related science, and network-based technologies are thus of central importance to maintaining and improving human wellbeing, as well as to improving economic, technological, and scientific competitiveness. Statistical physics offers a powerful set of concepts and methods to analyze problems of exactly the type posed by today's key challenges in network science. While the European statistical physics community has an established tradition of coordinated cross-border research collaborations in this area, there has been so far no significant European coordination effort on the initial training side. The proposed ITN is set to fill that gap. Its aim is to train a cadre of future research leaders in advanced methods of analysis, inference, control and optimization of network structure and dynamics, thus maximizing the impact of statistical physics approaches across a broad range of application areas. The ITN will create an innovative training environment by exploiting synergies and complementarities of a research matrix defined by methods and themes on the one hand, and application foci on the other. It will be implemented by targeted secondments of early stage researchers across both axes of the research matrix. Existing local training provisions and dedicated network-wide training events will be exploited to ensure that researchers are systematically exposed to the full range of statistical physics techniques and application domains covered by the 9 leading European research teams and 5 private sector partners integrated into this effort, and will comprise complementary skills training as an integral part. Keywords: statistical physics, inference, dynamics, optimization and control, systems biology, information technology, communications networks, financial networks, epidemic spreading, viral marketing 1 NETADIS, Background information for applicants Overview and overall objectives The Initial Training Network in Statistical Physics Approaches to Networks Across Disciplines (NETADIS) aims to train a cadre of future research leaders in advanced methods of analysis, inference, control and optimization of network structure and dynamics, to maximize the impact of statistical physics approaches across a broad range of application areas. Growing demand for expertise in this area derives from two main trends: (i) modern societies increasingly rely on network-based technologies, including mobile and land-line tele-communication, the internet or, lately, grid-computing, as well as on sophisticated transport and supply infrastructures – all contributing to rich and complicated dependencies that interlink virtually all modes of human activity; (ii) an ever widening range of complex problems at the frontier of research in biology, chemistry, economy, and social sciences are now being analysed in terms of network related techniques and paradigms. The present consortium will provide the training needed to enable a future generation of scientists to continue driving research at the forefront of these developments. Training will be organised via individual research projects, each situated in a frame of reference defined by (a) the principal research theme, characterized by the spectrum of methods and tools it employs and develops as well as by the type of problems it addresses, and (b) the principal application domain, defined by the nature of the system studied within the project in question. Principal research themes covered by the consortium are (i) inference of interaction networks from data, (ii) the analysis of dynamical processes on networks of given structure, and of network evolution, and (iii) optimization and control (of networks, network-evolution, and processes). Principal application domains of projects within the present ITN can be grouped into four areas: (i) systems biology and neuroscience, (ii) information technology, computing and algorithms, (iii) socio-economic systems and finance, and (iv) network phenomena in physical systems, specifically in laser physics. As described in greater detail below, there is a broad range of research challenges that will be addressed by the consortium. They include obtaining a systematic understanding of network reconstruction approaches, developing a framework for assessing reliability of inference of network properties from uncertain or incomplete state information, harnessing tools for the analysis of network dynamics required to capture essential complexities of specific application domains, and coping with scalability issues, i.e. with the need to develop robust algorithms able to efficiently handle the typically very large problem sizes in the various application domains. The principal conceptual framework to address these research challenges will be provided by statistical physics. Its central task has traditionally been to analyse macroscopic system behaviour in terms of properties and interactions of microscopic constituents, and it has developed a rich arsenal of concepts and techniques to efficiently deal with very large systems. More recently, neural networks and machine learning communities within statistical physics have been able to tap into and expand traditional tools and concepts of statistical mechanics and Bayesian statistics to successfully address a large class of inverse problems, dealing with the inference of properties and interactions of microscopic constituents from macroscopic system behaviour. Several members of the proposed training network have made pioneering contributions to many of the important advances that have harnessed concepts and tools of statistical mechanics to solve complex problems in network science, often outside the domain for which those methods and tools were originally conceived. Individual success-stories driven by members of NETADIS include (i) the derivation of the “survey propagation” (Braunstein, Mezard, Zecchina 2002/05) and “expectation propagation” inference algorithms (Opper and Winther, 2005) from the TAP and cavity methods of statistical physics, which are widely considered to be among the most powerful optimization algorithms to date; (ii) pioneering contributions to the field of econo-physics and finance (see Challet, Marsili, Zhang 2005 and Bouchaud, Potters 2003 for an overview), whose successful commercialization led to the creation of the CFM hedge-fund, a private sector partner of the network; (iii) the elucidation of collective dynamical processes on small world lattices (Monasson 1999); (iv) the efficient analysis of spectral 2 NETADIS, Background information for applicants properties of random graphs and matrices (Biroli Monasson 1999, Semerjian Cugliandolo 2002, Rogers, Pérez Castillo, Kühn, Takeda 2008); (v) the recent high resolution identification of protein complexes by combining bio-informatics and simulation methods (Schug, Weigt, Onuchic, Hwa, Szurman 2009). While members of the consortium have a proven track-record of research collaborations in the fields addressed by the proposed ITN, both ad-hoc and within previous European networks (including SPHINX, EVERGROW, STIPCO, DYGLAGEMEM, and PASCAL), there has been so far no significant European co-ordination effort in this area on the initial training side. This is a significant gap; the proposed ITN will fill it. Its key strengths include: It will bring together leading European research teams in the field of network analysis, inference, control and optimization. These teams will be able to provide a unified training framework for a broad range of methods that fall under the umbrella of statistical physics applied to network science. The ITN will generate innovation in research and training by bringing together researchers who are deploying statistical physics methods in a diverse range of application domains. Complementarity in training will be derived from coordinating research and training between teams working in the same application domain. It will be realized by organizing secondments of ESRs where they will be trained in complementary or related methods used by other teams working in the same domain. Synergies in training will be realized by organizing analogous secondments of students to teams working on the same research theme or with related methods, but in different application domains. These secondments to teams working both on different research themes and in different application domains serve a dual purpose: they will maximize the potential for new ideas to emerge, and will allow a new generation of researchers to integrate into research networks in a broader and deeper way than has so far been possible in traditional training environments. The associated private sector partners will provide complementary skills training in areas of particular relevance in entrepreneurial and non-academic environments. This will be implemented in the form of dedicated training sessions integrated into network-wide schools, conferences, and workshops, or as part of secondments to private sector partners where there is mutual benefit. The ITN will exploit the fact that networks are an area where scientific ideas and their application can be communicated to the public in a direct and intuitive way to support a significant level of outreach activities, with the aim of increasing public awareness of the importance of statistical physics as a research area with significant impact on everyday life. In helping to counteract an impression that physics deals mainly with black holes or colliders, it will contribute to shifting public perception of science more generally, and maximize the potential impact of the work of the generation of young researchers that we will train. High-level objectives for the network are: To train a cadre of future research leaders in advanced methods of analysis, inference, control and optimization of network structure and dynamics. To create a model for research training that crosses borders, disciplines and methodologies and that will both broaden and deepen the perspective of trained ESRs on the science they are exposed to. To foster a high level of direct collaboration between academia and industry by including specific project components identified as relevant for and beneficial to private sector partners. To enhance European leadership and competitiveness in areas of science and technology that lie at the heart of major socio-economic trends and emerging scientific paradigms. 3 NETADIS, Background information for applicants Scientific and Technological Objectives NETADIS will address four domains in the application of statistical physics techniques to the field of network science, and three broad themes that gather related issues across these domains. ESR training will be organized in the intersection of these application domains and themes, which form a research matrix. Each domain and each theme forms the topic of one training and research work package. Systems biology and neurobiology (WP4): The first sequencing of the human genome has been followed by an unprecedented acceleration in the amount and quality of genomic data. Beyond genomics, massive data sets are accumulating on transcriptomics, proteomics, metabolomics, and also, particularly relevant in the context of this work, on simultaneous measurements on many neurons (a field which does not yet have an established “omics” name). In all these fields there is the urgent need to extract useful and actionable information from mere data. Physicists, in particular statistical physicists, have been deeply involved in these efforts, in collaboration (and sometimes competition) with mathematicians, statisticians and computer scientists, as witnessed by the recent upsurge of interest in inverse statistical mechanics (inverse Ising/Potts models, described below at Inference (WP9)). At present numerous individual researchers and groups, typically with a strong background in statistical physics, and in several cases world leaders in their technical speciality, are moving towards and/or actively working in systems biology broadly defined, i.e. ranging from modern perspectives on classical biophysical subjects, e.g. cellular and tissue mechanics, to more molecular ones, such as regulatory and signalling networks and their evolution. The challenge that NETADIS will address, in this highly competitive and international field, is to scale up the endeavour by favouring the establishment of common training methods, collaborations and exchanges at the European level. Information technology and communication networks (WP5): The last decade has seen an unprecedented increase in the amount and richness of digital data which are collected, stored, analysed and acted upon, in all spheres of human activity. The key infrastructure that enables this process, as well as the internet of today, consists of communication networks, seamlessly mixing wired and wireless networks, with applications sharing real-time information and cooperating efficiently and reliably. This revolution in information technology, which has already profoundly changed our daily lives, is also a rich source of many interesting and challenging problems in statistical physics, which NETADIS will address. Network discovery starts from the fact that there is no central repository (or complete list) of links between ASes, the largest entities on the Internet, only empirically gathered collections (e.g. RouteView). Inferring the topology of such networks, as well as traffic, congestion and many other dynamic properties, from incomplete and locally stored data is an inverse problem sharing many of the features of network inference in Systems Biology (see above and below). Message-passing schemes, which have been widely analysed and developed in the statistical physics community, are methods of choice for such tasks. A further challenge is scalability. The effort required to monitor, control and allocate resources in a network grows with its size, and (in optimization and any other network task) researchers in NETADIS will aim to find algorithms where this effort grows only linearly with network size, i.e. in line with the available resources. For instance, the signalling overhead for computing and distributing even approximate solutions to network optimization problems can easily increase exponentially with network size, which is unacceptable. Computationally hard optimization problems are encountered at almost all layers of operation of today’s complex information and communications technology systems, and it is mainly through these optimization problems that statistical physics will make a difference. A further objective that NETADIS will address, and to which statistical physics has been shown to be especially well suited, is the task of modelling distributed systems in dynamic environments, where entities come and go and all operations are local. Energy efficiency and “green ICT” pose further challenges because they lead to optimization against competing criteria, e.g. the need to balance network performance against efficient energy utilization. Work done under the NETADIS umbrella will help here to devise distributed, 4 NETADIS, Background information for applicants computationally simple, low-processing-power methods that achieve close-to-maximal network performance. Finance and socio-economic systems (WP6): The importance of adequately modelling risk in networks of interconnected financial players has been highlighted in the recent financial crisis. Research on the spreading of defaults via direct infection through networks of financial exposures, known as credit-contagion, has begun only recently. Regulators have identified a number of key issues which require urgent attention and will be addressed within NETADIS. (i) The reliable inference of networks of dependencies from incomplete data. Central banks receive reports about interbank exposures only once these exceed a certain threshold. This limits central banks' abilities to assess economy-wide risk in financial systems, as does limited knowledge of direct dependencies that exist in the wider economy. (ii) Quantifying the role of credit derivatives, most notably that of credit default swaps (CDS). While in theory CDS should help to reduce credit risk, the recent financial crisis has shown that they can also be a source of risk in themselves, by creating additional contagion channels. (iii) Modelling the dynamics of illiquid markets. Addressing this problem is important for two reasons. First, the assumption of perfectly liquid markets, which underpins asset pricing and portfolio optimization theories, is badly aligned with common risk measures such as value-at-risk, which quantify expected losses, in situations of macroeconomic stress where markets tend to be illiquid. Second, the dynamics of illiquid markets has been recognized as creating an additional negative feedback channel which exacerbates the effects of credit-contagion. Several of these challenges that we will address have direct analogues in the domain of socioeconomic networks. To understand the dynamics of direct infection in social networks poses many scientific challenges due to the inherent heterogeneity of these networks. Progress on this issue, of obvious importance for health politics, will lead to a better understanding of the spread of epidemics, and more importantly will help to efficiently counteract this spread by suggesting optimal vaccination strategies. In the field of marketing a dual optimization problem is posed by finding marketing strategies that minimize dissemination efforts in advertising new products and services. The synergies offered by NETADIS will be exploited to address these challenges by creating and analyzing firstgeneration models of interacting asset prices, of credit-contagion models that include interactions created by CDS contracts, and of quantitative descriptions of infection dynamics, in a way that makes it possible to address the associated dual optimization problems. Laser physics (WP7): In recent years, random lasing materials (e.g. powders, porous media, precipitates in solution, or photonic crystals with impurities) have been extensively studied experimentally. Pumping energy into these systems causes them to re-emit multi-mode coherent light, with a spectrum displaying randomly arranged peaks in frequency. Starting from the structure and geometry of the atoms and molecules that scatter the light waves, one would eventually want a theory that predicts the nature of the modes: (i) where, or if, light modes localize, (ii) which spatial shape and size they have, (iii) how they overlap in space with nearby competing modes and (iv) on which frequencies they emit. The latter two features give the problem the structure of an interacting network of light-modes in a statistical mechanics representation. Indeed, a set of modes can interact only if their electromagnetic fields overlap in space and if the frequencies of the modes satisfy a modelocking condition. These rules strongly reduce the set of feasible interactions: each mode, viewed as a network node, can only have a limited number of connections, as in the other application domains above. A key challenge that we will address is the characterization of the structure of this network of modes and the strengths of the relevant random interactions, as is required e.g. in order to distinguish different physical regimes of lasers behaviour. The network has to be inferred starting from data acquired in measurements, of spectra and correlations of phases and amplitudes of the light modes, and this inference problem is closely analogous to those in our other application areas. There are three broad research themes across the above applications. The first of these is optimization and control (WP8). The correspondence between optimization and statistical physics has been exploited extensively since the invention of simulated annealing as a powerful method for solving complex optimization problems by way of physical analogy. Optimization problems have widespread applications in modern IT: in wireless digital communications, errors from transmission over noisy channels are corrected at the receiving end by iteratively solving combinatorial 5 NETADIS, Background information for applicants optimization problems in real time; practically all cell phones in use in the world today implement such schemes. Control means acting on information, hence changing the problem in response to external changes and internal processes like optimization. The connection between stochastic control and kinetic processes in physics (Langevin equations, Fokker-Planck equations) is by now very well established, and one of the main reasons why many physics graduates are employed as “quants” in finance. Control, and even more so collaboration and competition between different agents, are considerably harder problems than optimization per se, but ones where participants in the ITN have made important contributions. In our application areas, optimization and control (in the applied mathematics sense) are two of the main tools of systems biology; equipped with the arsenal of modern statistical physics, the ESRs we are going to train will be able to address larger and more complex problems in an innovative and robust manner. In information technology and communication networks we will bring the perspective of non-equilibrium stationary states to the quantitative description of distributed systems in dynamic environments. In finance and socio-economic systems, we will contribute quantitative understanding of the effects of limited (or noisy) data, e.g. the use of random matrix theory to estimate true correlations between assets. The second key theme is inference (WP9). The inverse problem is one of the most fascinating new issues to have surfaced in statistical mechanics in the past few years. It revolves around the possibility of inferring the interaction network (i.e. the couplings between degrees of freedom) in a large system of interacting units (e.g., genes or coherent light modes) from the empirical knowledge of certain statistical observables such as averages or correlations of the underlying microscopic variables (e.g. gene expression levels). Its relevance is widespread, from the study of biological or financial networks to the inference of the spatial overlap of localized light modes in random lasers. Finding fast and effective algorithms to solve the inverse problem is a central issue that NETADIS will address. This is reinforced by the fact that a standard description of large-scale complex systems, widely used in e.g. systems biology, is in terms of undirected correlation networks. However, correlation is not causation, as it may be mediated indirectly via intermediate components. It will be an important challenge to disentangle direct from indirect effects and to reconstruct the underlying causal network that generates the correlations observable in the data. There is a direct connection to the previous theme: many inference schemes such as maximum likelihood can be formulated as optimization problems. Consequently, network reconstruction often leads to NP-hard problems, where exact algorithms are restricted to small problems. But the relevant sizes of e.g. biological or social networks are huge. Just as in optimization and control approximate algorithmic approaches are needed, and this is where statistical physics can contribute. A concrete setting that has been studied intensively is inference using the maximum-entropy principle, which leads to the so-called inverse Ising/Potts problem: given a number of equilibrium configurations, is it possible to reconstruct the underlying generative model (in statistical physics language, the Hamiltonian of the system)? A specific feature of inference, as opposed to combinatorial optimization in general, is that the functional to be optimized (likelihood, entropy) is often not a local function, but depends on all the variables at once, as a partition function in statistical physics. The systematic local approximations of statistical physics have therefore proved to be highly competitive with techniques from e.g. the field of machine learning, and NETADIS will aim to develop these further. The third key theme is dynamical processes on graphs (WP10). This touches upon aspects from both the previous themes: a local search to solve an optimization problem is a (random) process on that graph, as are more complex distributed message-passing schemes, and understanding their performance from a systematic point of view is still (partly) an unsolved problem. Much more generally, most problems in all four application areas are naturally dynamic, and the static setting is only an approximation. The networks in systems biology change themselves through e.g. epigenetic modifications: the gene regulatory or protein-protein interaction networks active in different cell types can be very different. In peer-to-peer overlay networks as well as in wireless networks connectivity changes continuously, either by nodes entering or leaving the networks, or by moving around. In finance, time is essential. Even in inference, as described in more detail elsewhere in this project, it is important to go beyond the equilibrium setting of maximum entropy reconstruction, and to use nonstationary time series data as the input. No living system is stationary over very long times. In neural 6 NETADIS, Background information for applicants recordings stationarity holds at best on the time scales of hours, giving maximally tens of thousands of time points, and in global gene expression data the temporal range is many times less. Furthermore, important problems such as epidemics, economic or business relations are naturally understood as dynamic processes on static graphs. Statistical physics has known such processes for a long time, as (dynamic) percolation, Hopfield networks, Kauffman models, and the kinetic theory of spin systems. As is always the case, equilibrium statistical mechanics is, when it can be applied, much more powerful than non-equilibrium physics, but we will in NETADIS develop methodologies for both settings, including e.g. generating functional (path integral) techniques. The overall structure of the research programme envisaged is guided in its design by the structure of the challenges described above. Work on each application domain (Biology, IT, Finance, Physics) and each theme (Optimization, Inference, Dynamics), will be co-ordinated into one WP. These two aspects (application domain and theme) cross-link to form a matrix. The function of the WP structure is to provide efficient routes for training, for transfer of knowledge and ideas, and for enhanced research collaboration (a) within each application domain, thus exploiting complementarities between participants working in this area, and (b) within each theme, thus exploiting synergies in expertise between participants active in different application domains but tackling related scientific questions. Within the matrix formed by application domains and themes will be situated the individual research projects (IRPs) that the ESRs will work on. This is represented graphically in the table below. One of the key features of this proposal in creating an innovative research and training environment is that targeted secondments of the ESRs will take place along both directions of the matrix (a) “horizontally” within the application domain, to ensure that each ESR receives training in a broad class of themes and corresponding statistical physics tools and methods, (b) “vertically” across several application domains, to give each ESR an appreciation of the interdisciplinary nature of their work, and in particular the range of fields in which statistical physics ideas can be applied to networks. The private sector partners will make important contributions to the network. They will have input into the progress of all ESR research projects, e.g. via the network Supervisory Board, where all associated partners will be represented. They will also be represented on the Career Development Board, and so will be able to guide the training strategy for each ESR in the network, through the setting up and reviewing of career development plans. Private sector partners will also contribute directly to the network training programme, by providing courses for network-wide training and by hosting individual ESRs for targeted secondments. Importantly, the private sector network partners together cover the larger application domains represented in the research programme: The Human Genetics Foundation (HuGeF) will specifically have the remit to steer network training and research activities to maximize their relevance to biology, in particular human genetics. Collegio Carlo Alberto (CCA) will provide strategic oversight of all network projects with a financial or socio-economic dimension; it will also ensure that research efforts using statistical physics tools are benchmarked against the best practice of researchers in these fields. Capital Fund Management (CFM) will participate in guiding the research direction of ESR projects in the area of finance, and will deliver important training courses on Entrepreneurship, and on Exploitation of Research Results. Ericsson is linked into the network via a framework agreement with KTH ACCESS group, and this direct connection will provide useful input into projects in the area of communications and IT. Medialab will advise as to which application focus in each project will best lend itself to public outreach work. It will also be key to helping the network fulfil its public understanding of science mission, by providing network-wide training in media and outreach work, and advice to the network’s Dissemination Board. Control & optimization Systems biology and neurobiology Rome-1 Inference Dynamical processes KCL-1, Orsay-2, TUB, KCL-1, TUB Torino, Rome-1, NTNU NTNU 7 NETADIS, Background information for applicants IT & communications networks Orsay-1 KTH Orsay-1 KTH Finance and socioeconomic systems ENS, KCL-2 ICTP KCL-2, ICTP, ENS Rome-2 Rome-2 Laser physics Research programme matrix: Application domains (rows) and themes (columns) show where individual ESR projects are situated and how they connect along the horizontal (same application challenge) and vertical (same thematic challenge) directions. Note that individual projects can touch on several research themes, as indicated. As described above, the research programme is structured into four different application domains cross-linked with three themes, each domain and each theme co-ordinated into one research subprogramme. Each ESR will be seconded for 2-3 months to another participant working in a related application domain but using a different methodology (“horizontally”), and to another participant using a similar methodology but in a different application domain (“vertically”). In the remainder of this section we outline complementarities and collaboration between the participants and the associated private sector partners within the research sub-programmes horizontally, and summarize synergy and collaboration between the academic partners vertically. In the application domains, Systems biology and neurobiology groups together the participants KCL, Orsay, Berlin, Torino, Rome and NTNU. The Human Genetics Foundation (HuGeF, Torino), in close collaboration with Torino has a direct link to the research in NETADIS through its focus research area of statistical inference and computational biology, and brings in the dimensions of human systems biology and genomics/post-genomics platforms to the program.. Secondments to Torino will be extended to HuGeF, which has excellent wetlab facilities including next-generation sequencing equipment; HuGeF staff will also contribute to workshops through the training programme. IT and communication networks collects NETADIS participants Orsay and KTH. The telecommunication company Ericsson is an industry partner of the Linnaeus Centre ACCESS at KTH, which groups all the involved researchers from KTH. Secondments from Orsay to ACCESS will be extended to Ericsson when suitable joint projects between ACCESS and Ericsson are active. Finance and socioeconomic systems draws together participants ENS, KCL and Trieste. Collegio Carlo Alberti (CCA) and Capital Fund Management (CFM) are located in respectively Torino and Paris. CCA will host secondments when ERSs are working on projects where a wider economic theory and social science perspective is necessary; CCA staff will also contribute to workshops and outreach activities. CFM will host secondments to Paris, will give ESRs access to financial data, and will contribute hands-on knowledge of financial markets; CFM staff will also lecture at workshops and schools in the project. Laser physics as an application domain groups only Rome, since this is where the experimental data will be available. However, the entire network shares a background in statistical physics, which will be fruitfully exploited for vertical secondments. In the themes, control and optimization gathers participants Rome, Orsay, KTH, ENS and KCL, where the common underlying competence is combinatorial optimization, in particular large random combinatorial optimization problems. Inference links participants KCL, Orsay, Berlin, Torino, Rome, NTNU, KTH and Trieste and ranges through all four application domains. The common theme is finding an underlying (large) structure from (large) data, obscured by noise, under-sampling, or hidden due to the sheer complexity of the structures. The area of dynamical processes groups together participants KCL, Berlin, NTNU, Trieste and ENS. This theme joins together applications of statistical physics to domains outside physics, where on an abstract plane one is dealing with quantities on graphs that change in time. The private sector associated partners are for the most part not directly involved in the themes, these being more upstream in the development processes. An exception is 8 NETADIS, Background information for applicants CFM where staff have taken a leading role also in theory development and where we foresee intense contacts, primarily with Trieste, ENS and KCL, but also with KTH, Orsay and Rome. Training objectives As described above, the research programme is structured into four different application domains cross-linked with three themes, each domain and each theme co-ordinated into one research subprogramme. This section will describe how ESR training is organized to broaden and deepen the ERSs’ skills and knowledge, concerning both different approaches to addressing similar applications challenges (“horizontally”), and using the same or similar techniques in different application domains (“vertically”). The programme serves to address the need of integration of an overall broad but fragmented European effort. Systems Biology is a major area of modern Biology, the importance of which will only grow in the data avalanche to come out of next-generation sequencing. Statistical physicists made crucial early contributions to the subject, but the leading groups and the most attractive environments tend today not to be located in Europe. IT and communications networks provide crucial infrastructure for today's and tomorrow's world, and while Europe is the world leader on the wireless side, research ideas and new concepts are nevertheless more often developed elsewhere, cf. the recent success of the iPhone, and of Facebook and other social media. Here it is well recognized that central algorithms of modern IT (Belief Propagation, iterative decoding) have their counter-parts in statistical physics, and precisely in those areas (disordered systems, spin glasses) where Europe and in particular members of this consortium are leaders. In Finance and socio-economic systems, London is, in many areas, the world's centre and hub, but theories and new theory-rich ideas tend more often to be created in other parts of the world. Also here statistical physics has been important in providing what has proven to be fruitful training for students and researchers going on into a new area – this is witnessed (in the NETADIS consortium) by one of our private sector associated participants, CFM, as well as students from many of the groups now working in finance. Physical systems underlie all our infrastructure and potential for development and growth, but have, on the manufacturing side, partially moved out of Europe. One Marie Curie Initial Training Network cannot reverse the above long-standing situation and trends. But it can make sure that Europe's technical leadership in statistical physics of disordered systems is not dissipated into dispersed efforts, each group in the field seeking to collaborate with groups in an application area located nearby, and on those groups' terms, and that instead we can bring the expertise and skills of a whole integrated community to bear. NETADIS can leverage experience, knowledge and common language now spread out across many sites into an initial training effort which can start on a high level and proceed at an accelerating pace to bring a new generation of young European researchers to the highest level not only in their technical core area, but also in the application domains which are so important to Europe's future. On the workshop / postdoc training and dissemination of research, Europe is already on a high and integrated level, as witnessed by world benchmark programs (on this subject, several times) at Les Houches (organized by members of this consortium) as well as elsewhere in Europe, e.g. NORDITA, and even beyond (Spring 2008 and Spring 2011 programs at the Kavli Institute of Theoretical Physics China, organized by Aurell and others). This ITN addresses the stage just before, when students transform themselves into independent researchers and young colleagues, and from there eventually into research leaders. NETADIS aims to maximize the number of students who can make this transition as early as possible, and to translate as much of our technical lead in the themes as possible into progress and excellence in the application domains NETADIS will fund a significant public outreach activity. Explaining science, and especially theoretical science, is not easy. Within NETADIS we will have an exchange of ideas and material (including presentation material) as well as skills (including presentation skills) to help individual 9 NETADIS, Background information for applicants researchers to reach out to the public in their local communities and on their national level, to promote the public's understanding of our science, and of science as a whole. The overall mission of the initial training network will be reached with a combination of training for individual ESRs, and network-wide training. Each of these channels will deliver both scientific skills (key methods of statistical physics; awareness of breadth of application domains) and complementary skills. A central plank of the implementation of this system is a career development plan for each ESR. Training for individual ESRs will consist of the following elements: Each ESR will benefit from on-the-job training obtained from working with the leading research groups that this network gathers. Each ESR will be seconded for 2-3 months to another network participant working in a related application domain but using different methodology or addressing a different research theme. These secondments will broaden the methods and skills base of the ESRs. Each ESR will have one further secondment of similar or somewhat shorter length, to a network participant using similar methodology but in a different application area. This mechanism will strengthen the awareness of ESRs of the range of application domains, and will foster a spirit of looking beyond the immediate research project to complementarities and new ideas from other domains. The timing and choice of both secondments will be targeted so as to take maximum advantage of the wide range of local training opportunities in the participant and partner organizations; some key examples of these opportunities are described in the next section. Where possible, secondments will be to the private sector (HuGeF, CCA, CFM, Medialab, and Ericsson via KTH/ACCESS). Practical training in management skills will be given by involving all ESRs in the organization of network activities such as schools, general network meetings, and workshops. All ESRs will also be encouraged to take full advantage of opportunities for local complementary skills training. A broad range of options is offered in all participating host institutions. Network-wide training activities will comprise: A scientific kick-off meeting once all ESRs have been recruited and career development plans drafted locally. At this meeting, each ESR will present their own career development plan to all network members, including the Career Development Board. Career development plans will be reviewed and modified as necessary in light of feedback and suggestions received. An added benefit of the career development plan presentations will be that the scheme will fix the idea that it should be the ESRs that give talks at network meetings. In this way, the training benefits in presentation skills that accrue for the ESRs are maximized. A dedicated training session on this topic will also be organized at the kick-off meeting. Two schools will be organized, taking place roughly 8 and 20 months after the kick-off meeting. These will each focus on two scientific themes, but cover several application areas. We envisage (1) Inference/machine learning and optimization, covering applications in bio-informatics, molecular & systems biology, neuroscience, IT and communication, and laser physics; and (2) Dynamical processes, covering applications in systems biology and neurobiology, in finance, social networks and marketing. The precise content will, however, be adjusted strategically by the Education Committee to best meet the training needs of the ESRs, as identified within their Career Development Plans. The network participants in Berlin, London and Trieste have already agreed to host schools. The schools will integrate a substantial component of complementary skills training, delivered by the private sector partners. The Trieste school will include a dedicated course on science communication and outreach, run by Medialab. The other summer school will have speakers from CFM giving courses on entrepreneurship, and exploitation of research results. Together with presentation skills training, we consider these aspects to be the most important for 10 NETADIS, Background information for applicants MonTh 1-3 4-6 7-9 10-12 13-15 16-18 19-21 22-24 25-27 28-30 31-33 34-36 37-39 40-42 43-45 46-48 enhancing the ESRs employability and career prospects. We envisage also possible contributions from CCA and HuGeF on project management and patenting, and will explore the possibility of bringing in additional commercial sector training expertise from the Compagnia di San Paolo bank foundation, the largest bank foundation in Italy, via its close ties to CCA and HuGeF (which are Operating Bodies of the foundation). Private sector network partners will also contribute to the scientific content of the schools, e.g. HuGeF and CCA could offer short courses in bio-informatics and aspects of economics and statistics as part of the schools. Eminent external speakers from academia would be invited to give keynote talks or longer tutorial sessions, and candidates for this have already been identified (V. Balasubramanian, W. Bialek, J. Carbonnel, M. Cieplak, P. Domingos, J. Hertz, M. Jackson (Stanford), B. Kappen, T. Mitchell, B. Schölkopf, M. Seeger, M. J. Wainwright, O. Winther) A final conference for the network will be run towards the end of the network period. This will bring together the research results and give an opportunity to plan applications for follow-up funding that would maintain and enhance the achieved integration of training and research activities. It will also act as a show case towards end users (see section on dissemination below). ESRs will also be expected to attend the Spring School of the International Master’s Programme in Physics of Complex Systems, which is described in more detail below and provides training that is highly relevant to the themes of the network. KCL 1 KCL 2 ENS Orsay 1 Orsay 2 TUB Tori- Rome no 1 ESR Recruitment Rome 2 NTNU KTH ICTP Scientific kick-off meeting & workshop; complementary skills course: presentation skills International Master's Spring School (option 1) TUB ICTP Medi alab KTH HuGeF NTN U Rome KCL Torino ENS Orsay CFM First network school & complementary skills course, incl. entrepreneurship & exploitation of results International Master's Spring School (option 2) Trieste CCA Orsay ENS Rome KCL KTH HuGeF NTNU ICTP Torino TUB Second network school & complementary skills course: communication and outreach Final network conference Table of training schedule for individual ESRs, showing key network-wide training events as well as (in italics) secondment periods. Timing and destination of secondments is indicative and will be finalized in individual Career Development Plans. Smaller workshops (1-2/year, not shown) will be targeted at specific ESR training needs. Link to International Master’s programme: Members of the Orsay team are involved in an International Master’s programme in Physics of Complex Systems, together with the Politecnico in Torino and SISSA and ICTP in Trieste. This Master’s (http://www.polito.it/pcs) is centred around the concepts and techniques of theoretical and statistical physics, and open to interdisciplinary applications in computer science, biology and finance. It is a two-year programme that in its second 11 NETADIS, Background information for applicants year includes a spring school at SISSA/ICTP in Trieste, open to external students at the Master’s and PhD level, which has the dual scope of exposing the students to contemporary research in interdisciplinary statistical physics and of helping to foster professional contacts and interactions between the students and senior researchers. The ESRs of the network would have privileged access to this school, and would be expected to attend it in either Spring 2013 or 2014. Apart from the Spring School, the Master’s programme is organized into four semesters: A first semester in Trieste at SISSA and ICTP for Italian students and in their home universities in Paris for the French students (courses in French); international students have the option to choose between the two alternatives; A second semester in Torino at the Politecnico di Torino; A third semester in Paris at a consortium involving Universities Pierre & Marie Curie (Paris 6), Paris Diderot (Paris 7), Paris-Sud (Paris 11) and the École Normale Supérieure at Cachan. Finally, the fourth semester is dedicated to the European multidisciplinary Spring School described above – where students can choose between various short courses – and to a research project. All the phases of this programme will be open to participation by ESRs from the NETADIS training network. 12