CEDRO falls into the broad theme of performing decentralized inference (stochastic optimization, estimation, and learning) over graphs. It notably recognizes the increasing ability of many emerging technologies to collect data in a decentralized and streamed manner. Therefore, the focus is on designing decentralized approaches where devices are collecting data in a continuous manner. The project also recognizes that modern machine learning applications (where tremendous volumes of training data are generated continuously by a massive number of heterogeneous devices) have several key properties that differentiate them from standard distributed inference applications. Particular focus will be given to developing and studying approaches for decentralized learning in statistical heterogeneous (multitask) settings in the presence of limited communication resources and heterogeneous system devices. The project emphasis will specifically be on illustrating the interest of the proposed approaches in machine learning frameworks using publicly available datasets.

By 2021, cloud IP traffic will be the most part of an Internet traffic that complexifies with an increasing devices diversity and traffic dynamicity. A proposal framed at the cloud to face this situation is the Knowledge Defined Networking (KDN), where Machine Learning (ML) and Artificial Intelligence (AI) are combined with SDN/NFV and network monitoring to collect data, transform them into knowledge (e.g. models) via ML, and take decisions with this knowledge. Under this paradigm, we aim to design a unified AI-based framework able to learn new efficient cloud network control algorithms. This framework will integrate seamlessly data-driven control (based on ML tools) and model-driven control (based on optimization models), addressing scalability and optimality issues of the cloud control. To do that, we intend to apply two promising AI tools: Deep Learning (DL); and, Reinforcement Learning (RL). In the project, a Deep Learning Artificial Neural Network (ANN) will be used to transform the original input data representations (in our case, the cloud network state) into a low dimensional space where the network structural information and network properties are maximally preserved, and used them to solve in a more tractable way the optimal control problem. RL will be applied to learn the optimal control by interacting with the environment (in our case, the Cloud network).These interactions can be used to guide the learning of the weights of the deep ANN. The result is that the RL algorithm (acting as control loop) will solve more easily the control problem using as input this more compacted and lower dimensional representations found by the deep neural network. The main novelty of our approach is that we state that, for network control problems, the deep ANN should not be implemented using the same deep layer architectures used in computer vision (the so-called convolutional layers), but using a different kind (the so-called novel graph embedding architectures) better fitted to the graph nature of the network problems. Then, we propose to use the graph embedding layers as deep layers to solve cloud network control problems, namely the dynamic allocation of service chains composed by network virtualised functions. Starting from the case where the network service is unicast, we will move later to the multicast case, since video delivery, the classical multicast service, is the Internet killer application. Finally, we will implement a KDN proof-of-concept tested where our Deep Reinforcement Learning control will send via the northbound interface the control decisions to a an SDN controller, that, in its turn, will instruct an emulated SDN network.

Rheumatoid arthritis (RA) is a chronic autoimmune disease affecting approximately 1% of the population, and is characterized by joint inflammation leading to structural damage and disability. RA often exhibits variable disease activity over time with exacerbations (relapses) and periods of low disease activity. RA patients are currently heavily treated by administration of pharmacological drugs or biologics. In addition to compliance and side-effect issues of these drugs, relapses occur in one third of cases. While rheumatologists assess clinical symptoms of RA and try to adjust treatment in case of relapse or adverse side effects, they only intervene when symptoms worsen due to the lack of reliable point-of-care biomarkers for RA relapses. This likely results in sub-optimal therapeutic drug levels. A relapse-responsive drug delivery system is highly desirable because it would titrate drug release to match the disease activity that would notably allow personalized medicine to be developed, tailoring therapy to the individual, shortening time from onset to effective treatment, improving cost and risk-benefit ratios of drugs, and ultimately achieving high response rate with minimal toxicity and/or desensitisation. In parallel of the pharmacological approach, Bioelectronic Medicine is using electronic devices to modulate the activity of the human autonomic nervous system. This new approach has been tested with some sucess in PR patients. However, not only less than one third of patients achieved real remission following electrostimulation but, some patients also exhibited adverse effects. Both pharmacological and bioelectronic approaches would benefit from personalized-treatment through the anticipation of relapse phases for early treatment. One approach to improve efficacy and reduce adverse effects could be to deliver on-demand quality-controlled electrostimulation based on information extracted from neural signals using advanced signal processing, control and machine learning approaches. The aim of this project is to provide the proof-of-concept in mice that sympathetic neural activity can be used as a new class of biomarker for a personalized automated real-time monitoring to prevent relapse in arthritis. Pertaining to this proof-of-concept, our specific objectives are: • Objective 1: To identify the signals that generate neural activity changes in splenic nerve following inflammation; • Objective 2: Building on this neural signature, to identify an electrical fingerprint in the splenic nerve that predicts relapse in a mouse model of inflammatory arthritis; • Objective 3: To investigate whether therapeutic electrostimulation can prevent relapse when delivered automatically following detection of electrical fingerprints identified in the previous objective (closed-loop system). The main innovation of this project relies on the identification of real-time and reliable biomarkers based on machine learning methods that could discriminate neurograms associated with the relapse periods in RA patients to anticipate therapeutic intervention. To achieve our goal, we have assembled a multidisciplinary consortium of two academics partners, which are both internationally recognized leaders in immunology and biomedical signal processing fields. We also have obtained preliminary results suggesting that spontaneous splenic nerve activity could be used as a biomarker for inflammation. Combining disease biology with advanced analysis on neural decoding will be the strength of the consortium. This approach will allow us to establish the scientific foundation of a new field of research at the interface between neurology and immunology that may be beneficial for patients suffering not only from RA but also with other Immune-mediated inflammatory diseases including Crohn’s diseases or Multiple Sclerosis.

This project will make a dual contribution inherent in its specific dual character, both civilian and military. At the civilian level, the research will allow us to better understand the argumentative strategies used in controversies on international issues and military operations. At the military level, the work carried out will enable us to better understand the information environment of military operations and its mechanisms. The proposed research is organized into two Axes. To understand the impact of contreversaries on armed conflict, the first axis studies the controversies in English and French taking place in the conflict between Russia and Ukraine. The second axis will focus on designing and implementing an Artificial Intelligence algorithm for the automatic analysis of argumentation in these controversies. This work is based on the joint creation of an annotated dataset from an initial mapping, allowing automatic argumentation analysis to identify and classify controversies. Building on the practical field of the conflict between Russia and Ukraine, this project will lead to the creation of a decision support tool, capable of mapping the actors through the identification of controversies and their characterizations by means of argumentation and counter-argumentative strategies. The entire research project will broaden our knowledge on the use of artificial intelligence to study controversies and identify arguments for a better knowledge of the information environment of operations.

Several recent revolutions in imaging rely on numerical computations. One can think of single molecule localization microscopy (Nobel Prize 2014) or cryo-electron microscopy (Nobel Prize 2017). What they have in common is the need to perform prior mathematical modeling and calibration of the system. Although they have made it possible to observe phenomena that were previously out of reach, their expansion is currently limited by an important problem: it is difficult to precisely control the imaging conditions (e.g. temperatures, wavelengths, refractive indices). This results in modeling errors that can have disastrous repercussions on the quality of the images produced. Thus, these technologies are currently reserved for a handful of research centers possessing state-of-the-art equipment and considerable interdisciplinary experience. The objective of this project is to bring new theoretical and numerical solutions to overcome these difficulties, and then to apply them to different optical microscopes. This should allow to democratize their use, to reduce their cost and the preparation time of the experiments. The central idea is to characterize a measurement device, not by a single operator (e.g. a convolution), but by a small dimensional family allowing to model all possible states of the system. To our knowledge, this idea has been very little explored so far and opens many difficult questions: how to evaluate this family experimentally and numerically? How to identify the state of the system from indirect noisy observations? How to exploit this information to reconstruct images in short computing times? We have begun to explore these questions in recent works and wish to continue this effort using tools from optimization, harmonic analysis, probability and statistics, algebraic geometry, machine learning and massively parallel computing. We hope to make significant advances in the field of blind inverse problems. We will validate them on photonic microscopy problems in collaboration with opticians, responsible for two microscopy platforms. This will allow us to obtain direct feedback for real problems in biology. We will particularly study the problems of super-resolution by single molecule, multi-focal localization and blind structured illumination. Moreover, several companies in the Toulouse area (INNOPSYS, IMACTIV-3D, AGENIUM), will provide us with data from their microscopes (line scanning microscope, light sheet fluorescence microscope), which will ensure direct transfers to industry. The impact of this project goes far beyond optical microscopy, since similar problems exist in astrophysics, earth observation, non-destructive control and seismology. The funding of this project would promote the creation of an interdisciplinary team in computational imaging at the Center for Integrative Biology in Toulouse (CBI) and would encourage interactions and knowledge exchanges between the MORPHEME team in Sophia-Antipolis who is already a reference in the Côte d’Azur region and the new team in Toulouse.