Populations are increasingly vulnerable to disastrous natural or technological events, as demographic and urban growth lead to greater exposures of goods and people. Large scale evacuation strategies are efficient tools for mitigating this vulnerability. Nonetheless, risks incurred during an important displacement through an altered environment are high: refusal to evacuate, crashes, direct exposure to the source hazard, riots, emergency services failures… In France a policy called Territoires à Risques importants d’Inondation (TRI) has emerged to deal with floods, in a first step to deal with the most frequent natural disaster in this country. Nevertheless, local governments and emergency managers lack prospective tools to assist their understanding and planning of large scale evacuations. ESCAPE aims at overcoming this major problem by the creation of an evacuation operational research system. The core of our project is the tight coupling between Geographical Information Systems, agent-based multiscale modelling and computer simulation exploring tools. It will be deployed and validated on real case studies, so as to generate simulations realistic enough to allow their use by emergency managers for experimenting evacuation strategies. By combining sources including territorial information (land occupation, transport networks, hazards expansion and intensity), demographic data (residential and transitional population numbers, age pyramid), a mobilityand traffic management simulator (cars, bikes, pedestrians, public transport), and by providing different evacuation strategies (partial or complete, by waves or synchronous), we will provide measures on evacuation time of various crisis zones, and will make explicit local and global constraints on these times. For that, we need to explore at multiple space and time scales the emergence of collective behaviours that would detract from planned strategies, and to devise solutions to dampen the consequences of these behaviours on the evacuation times. The ESCAPE team will build demonstrators to allow productive interactions with emergency services and remain reality-grounded for the whole duration of the project. These prototypes will allow us to precisely identify the stakes at play in each case study and the needs of the various managers.

Water transport and distribution systems must be carefully monitored and operated to avoid water losses, to safe energy and to protect the assets of the water utilities against damages. In major water transport and distribution systems this task is performed by centralized SCADA (Supervisory Control and Data Acquisition) systems which receive information from remote sensors and remotely control components like valves and pumps. SCADA systems are technically complex and expensive. For this reason, they often are not achievable for small water utilities which do have to operate their transport lines and distribution networks manually. Recent developments in ICT (Information and Communications Technology) open new paths for technologically advanced, low-cost solutions for water system monitoring and control. In the first place it is the decentralized IoT (Internet of Things) approach which already gained significant importance in the industrial sector but is still in an embryonal stage considering water utilities. To explore the promising potential of IoT technologies in combination with other innovative ICT the IoT.WATER project sets up an interdisciplinary consortium, bringing together the expertise needed for this task: •The Institute of Fluid Mechanics and Turbomachinery of the Technical University of Kaiserslautern, Germany (SAM) is specialized in optimization of machine design, condition monitoring and decision support for optimized operation of hydraulic and thermal turbomachinery. SAM will act as project coordinator. •The Centro de Pesquisas Hidráulicas e Recursos Hídricos of the Federal University of Minas Gerais, Brazil (CPH) is a hydraulic research center dedicated to support the energy and water supply sectors, with a team of professors on the fields of civil, mechanical, electrical and automation engineering. •The research group HECE of Liege University conducts experimental and numerical research in hydraulic engineering. The group has been developing the modelling system WOLF, which enables the computation of pressurized, free surface and mixed flow in channel and pipe networks. The flow models are coupled to self-developed optimization algorithms. WOLF is routinely used for teaching, research and consultancy. •The MIND research group from INSA Rouen Normandie, France (MIND) conducting research in the fields of Multi-Agent Systems, Semantic Technologies and Human-Machine Interactions. Focus is the study and development of interaction and decision-making processes in mixed communities or in cyber-physical systems. •The Dr. Kraetzig Ingenieurgesellschaft mbH, Aachen, Germany (KI) as a SME engineering company with expertise in setting up distributed monitoring and control networks for environmental control, water system optimization and water loss reduction. The project proposal follows a systematic approach starting with high time resolution measurements in water systems (already based on IoT technology), numerical description and characterization of the water systems, adaptation and development of hydraulic models suitable for near-real-time simulation, development of tools for implementation in IoT nodes for optimizing of component operation (e.g. pump operation / condition monitoring) and for supporting decision making (e.g. energy efficient management, alarm generation in case of incidents etc.), testing and system comparison by use of “water system digital twins” and physical water system models. The overall IoT system will be installed in systems in Belgium, Brazil and Germany for extended field testing. Further application in other water utilities will be supported by documentation of the used methodology. The outcome of the project will be a new ICT approach for low-cost water system monitoring and operation. This system will contribute to water loss reduction, backing the efforts against impacts of water scarcity, simpler maintenance strategies and strengthening of water systems resilience against havocs.

Since the Brundtlant report in 1987, the assessment of the sustainability of human economic systems has moved from a vague concept to a consistent set of methodologies and modelling tools within the area of Sustainability Science. These methods are nowadays applied in most of the economic sectors, both in industry and in policy making, and are developed in academia worldwide, one of the major outcomes being the eco-design of human activities. This is more particularly the case of Life Cycle Assessment (LCA), ruled by the ISO standards 14040-44 and at the core of the EU policy making and scientific research. After 20 years of steadily development, efforts are currently oriented toward the deepening and broadening of LCA, to support the development of more environmentally sound products for consumers as well as to steer policy making on key societal issues, like e.g. electromobility, building and construction and biotechnologies. Among these developments, the introduction of time dependency in LCA has been dramatically underestimated and underexplored. In conventional LCA, human driven systems are typically considered to run in steady state conditions, including fully elastic activities, neglecting time lags and stocks of goods and products. At best, the current practice considers different scenarios (related to time horizons) where relevant inventory parameters (e.g. related to the production functions like electricity production) are changed according to possible technological, market and regulatory evolutions. Following the same line of reasoning, the lifecycle impact assessment models translating the inventory results into environmental impacts have limited coverage and consideration of dynamic features (related e.g. to pollutant fate, exposure and effect parameters). Conventional LCA models are indeed ideal simplifications of a reality which is, nonetheless, highly dynamic and variable over time. The main objective of DyPLCA is to develop a comprehensive and operational approach (methodology and tools) for the proper consideration of time dependency in LCA, with strong emphasis on the development of an integrated modelling solution for both the life cycle inventory (LCI, at foreground and background levels) and the life cycle impact assessment (LCIA) phases. Results at the end of the project will be a methodology, models and computational tools for true dynamic LCA, well beyond the current practice based on forecasted scenarios, in a form readily usable for LCA practitioners. The modelling framework will be tested and applied to three relevant test bed LCA applications: 1) bio-technologies (complex non steady state, cyclic functioning); 2) buildings (long term non steady state); 3) car tire traffic noise (highly dynamic and stochastic). These systems were selected because of their contribution to the overall environmental impacts generated by human driven economies as well as because of the pertinence of the temporal scale in the assessment. DyPLCA will provide new scientific knowledge, clearly beyond the current state of the art of the science of LCA, focusing in particular on 1) the deepening and broadening the scope and modelling of LCAs in an rather unique way, through the combination of temporal characterization techniques and LCA and the harmonization of micro-process level inventories (i.e. Ecoinvent v3 datasets) with time behaviours of large scale systems; and 2) full implementation of these modelling and investigation approaches on three practical application situations of broad societal interest. Apart from the scientific communication and promotion of scientific and technical culture, higher education will benefit from the project results, thanks to the academic partners involved, and the repercussions on the respective research strategies of the project partners are huge.

In recent years we have witnessed an explosion of successful applications of deep learning including speech recognition, automatic translation, self-driving cars, computers that can beat professional Go players, and recommender systems. Deep networks designed for these tasks have millions and billions of parameters that take enormous resources to train in terms of data, memory and computing power. Deep networks are clearly data, computationally and memory intensive, making them difficult to use and to deploy in particular on embedded systems. Applications in embedded devices ---such as smart phones, tablets, but also intelligent vehicles such as self-driving cars and drones, as well as wearable devices--- require low-power and memory efficient solutions to to solve recognition and scene understanding problems. Such requirements are not well aligned with the current resource-heavy approach to deep learning. Reducing the number of parameters and computational requirements, while preserving predictive performance, is critically important for deploying deep networks in this context. Deep in France collaborative research program aims at expanding the frontier of green deep learning. Green deep learning refers to the practice of using deep learning more efficiently while maintaining or increasing overall performance. Our vision is to develop theory and new deep learning architecture, algorithms and implementation allowing to deal with limited resources in terms of training examples and computing power. It will facilitate widespread use of deep learning to yet under-explored application domains such as audio scene recognition, embedded perception and video prediction. The bottlenecks regarding deep learning addressed by \projectname are both theoretical and methodological, as well as technical related to the optimization procedure and implementation. Our project also aims at bringing together complementary machine learning, computer vision and machine listening research groups working on deep learning with GPU’s2 in order to provide the community with the knowledge, the visibility and the tools that brings France among the key players in deep learning. The long-term vision of Deep in France is to open new frontiers and foster research towards algorithms capable of discovering sense in an automatic manner, a stepping stone before the more ambitious far-end goal of machine reasoning.