The goal of the project is to develop energy-efficient and scalable designs for distributed mobile networks operated by renewable energy sources, and with very limited signaling overhead and computational complexity. The project will design wireless networks, composed by smart nodes, which are able to perform the following tasks: 1) Self-configure, autonomously allocating their own physical layer radio resources through energy-efficient, feedback-aware, and complexity-aware radio resource allocation. 2) Self-sustain by harvesting energy from renewable and intermittent energy sources, such as solar and wind, as well as from dedicated radio frequency signals present over the air. 3) Sharing or trading energy with other network nodes in order to prolong the lifetime of nodes which are low on battery and to obtain a more fair energy distribution across the whole network. The long-term vision of the project is to kick-start a paradigm shift from core-centric, throughput-optimized networks, towards device-centric, energy-optimized networks. The development of autonomous, energy-independent, and self-organizing wireless networks will enable: a) 100% coverage in urban environments in a power efficient manner; b) network coverage in remote/developing areas where currently it is commercially unattractive to do so. These advancements will make the vision of a connected society sustainable, almost zeroing the operational expenses related to power and hence revolutionizing the business models for ICT by radically reducing energy costs. This will also facilitate the rise of new markets and applications in remote areas, which contributes to closing the digital divide between densely populated and scarcely populated or rural areas.

Deep learning is an enormously successful recent paradigm with record-breaking performance in numerous applications. Individual autoencoders (AEs) of a multilayer neural network are trained to convert high-dimensional inputs into low-dimensional codes that allow the reconstruction of the input. Although some explanations appear to be solidly grounded, there is no mathematical understanding of the AE learning process. This project is a collaborative endeavor of researchers with strong complementary backgrounds. Its main innovation is the idea to capitalize on powerful and fertile concepts from information theory (expertise of researcher) in order to advance the state of the art in deep learning (expertise of supervisor at TC). The innovative research work is motivated by our recent insight that there is an intimate relationship between AEs, generative adversarial nets and the information bottleneck method. This method is a model-free approach for extracting information from observed variables that are relevant to hidden representations or labels and will serve as basic building block for an information theory of representation learning. The planned objectives are split into 3 workpackages: 1) information-theoretic criteria and statistical tradeoffs for extracting good representations, 2) structured architectures/algorithms for learning, 3) use of stochastic complexity to assess the descriptive power (model selection) of deep neural networks. Accomplishing the challenging goals of this proposal requires a variety of methodologies with a rich potential for transfer of knowledge between the involved fields of information theory, statistics and machine learning. Our new framework is expected to bridge the gap between theory and practice to facilitate a more thorough understanding and hence improved design of deep learning architectures. The fellow researcher is coordinating the LIA Lab of the CNRS (started in 2017) where he is collaborating with the supervisor at TC

To fight climate change, we must urgently reduce the CO2 emissions caused by fossil-fuel combustion, which represents today over 80% of the primary energy production. Clean electrified solutions are on the horizon but are unlikely to reach commercial development before 2040. Novel CO2-neutral (biofuels) or CO2-free (H2) combustion technologies are widely considered, but these technologies face increasingly stringent regulations on pollutant emissions, in particular nitric oxides and carbon monoxide. To reduce pollutants, the strategy is to use low-temperature flames. However, these flames are prone to instabilities and extinction, thus causing safety issues. Plasma-assisted combustion (PAC) is a highly promising method to stabilize low-temperature flames thanks to the extraordinary ability of plasma discharges to efficiently produce combustion-enhancing radicals. Today, however, their effects on pollutants are poorly understood and their scalability to industrial combustors remains to be proven. Our goal is to bring PAC to the level of maturity needed to make it practical on real combustion devices. For this, we will first elucidate the thermochemical mechanisms of plasma stabilization in CH4- and H2-air flames and their impact on pollutant emissions. This will require measuring the rates of poorly known reactions involving excited electronic states of molecules with advanced femtosecond optical diagnostics. With this knowledge, we will explore two novel strategies to minimize pollutants. We will then develop a robust and versatile multi-physics model to predict PAC effects in large-scale combustors. The final challenge will be to demonstrate for the first time a stable, low NOx, hydrogen/air flame in a combustor representative of aircraft engines. Beyond combustion, this project will open novel ways to better predict, control, and enhance chemical processes in applications such as hydrogen production, CO2 conversion, bio-decontamination, or materials synthesis.

Mobile data traffic sharply increases each year, due to the rich multi-media applications, video streaming, social networks, and billions of connected users and devices. This increasing mobile data traffic is expected to reach by 2018 roughly 60% of total network traffic. In this regard, caching contents at the edge of the network, namely at the base station and user terminals, is a promising way of offloading the backhaul (especially crucial in dense network deployments) and decreasing the end-to-end content access delays, since the requested contents become very close to the users. Therefore, caching has the potential to become the third key technique for wireless systems sustainability. The goal of this proof of concept is to realize a prototype of such an architecture which enables caching at the edge of the network, and called as “CacheMire”. In particular, we shall focus on development of the first version of CacheMire, which aims to 1) provide an application programming interface (API) to website developers (or content providers); 2) build a set of software/hardware tools to track/collect users' content access statistics under privacy constraints and regulations; 3) and design a storage unit/box for caching strategic contents (i.e., images, videos, files, news) at the base stations and access points. In addition, we aim to combine advanced physical layer techniques with caching so that resources in the uplink/downlink of next generation 5G wireless networks can be further optimized.

The climate change is now one of the biggest threats faced by our natural world and leads to many extreme weather events, flooding, wildfires and heatwaves. Unmanned Aerial Vehicles (UAVs) are the promising technologies for the emergency response applications (ERAs) to provide more efficient and faster services to save lives and reduce economic loss. This project aims to develop an innovative cooperative, connected and intelligent UAVs (CIUs) for ERAs, where sensing and computing resources, and flying information from individual UAV could be shared and exploited through effective communications and control of the CIUs. Specifically, context-aware and service-oriented UAV-to-everything (U2X) networks; robust cooperative UAV sensing and computing (CSC) schemes and intelligent cooperative UAV control strategies for ERAs will be investigated. An international inter-sector and inter-disciplinary consortium consisting of world leading academic institutions and prominent industrial partners is created to collaborate on developing novel CIU technologies in ERAs. Cutting-edge communications networks, edge computing, cooperative sensing, intelligent control, and machine learning related technologies will be investigated to tackle the associated challenges. With competent and complementary expertise of the partners and their extensive international research collaboration experience, this project will promote research innovation, foster knowledge sharing, enhance the potentials of participating researchers, and contribute to the European leadership in the UAVs, cooperative sensing and computing, multi-agent control, information and communications technology (ICT), and emergency response sectors. The developed CIU technologies and applications are not only applicable in various ERAs, like flooding, wildfires, and earthquake; but can also be readily extended to other applications, such as smart cities, public safety, agriculture, and the wider scientific community.