New generations of mobile access networks promise low delay and high-speed throughput data connections paired with in-network processing capabilities. IoT data and local information available to users’ devices will feed AI-based applications executed in proximity on edge servers and service composition will routinely include such applications and their microservice components. PARFAIT tackles new resource allocation problems emerging due to the need of distributed edge orchestration of both computing and communication, in a context where the unknown footprint of AI-based applications requires advanced learning capabilities to permit efficient and reliable edge service orchestration. The PARFAIT project develops theoretical foundations for distributed and scalable resource allocation schemes on edge computing infrastructures tailored for AI-based processing tasks. Algorithmic solutions will be developed based on the theory of constrained, delayed, and distributed Markov decision processes to account for edge service orchestration actions and quantify the effect of orchestration policies. Furthermore, using both game and team formulations, the project will pave the way for a theory of decentralized orchestration, a missing building block necessary to match the application quest for data proximity and the synchronization problems that arise when multiple edge orchestrators cooperate under local or partial system view. Finally, to achieve efficient online edge service orchestration, such solutions will be empowered with reinforcement learning techniques to define a suit of orchestration algorithms able to at once adapt over time to the applications’ load and cope with the uncertain information available from AI-based applications’ footprints. Validation activities will be designed to demonstrate real-world solutions for practical orchestration use cases, using both large scale simulation experiments and research testbeds.

5G networks are expected to revolution our living environments, our cities and our industry by connecting everything. 5G design has, thus, to meet the requirements of two “new” mobile services: massive Machine-Type Communications (mMTC), and Ultra Reliable Low Latency Communications (URLLC). Slicing concept facilitates serving these services with very heterogeneous requirements on a unique infrastructure. Indeed, slicing allows logically-isolated network partitioning with a slice representing a unit of programmable resources such as networking, computation and storage. Slicing was originally proposed for core networks, but is now being discussed for the Radio Access Network (RAN) owing to the evolution of technologies which now enable its implementation. These technologies include mainly the tendency for virtualizing the RAN equipment and its programmable control, the advent of Mobile Edge Computing (MEC) and the flexible design of 5G on the physical and MAC layers. However, the complete implementation of slicing in the RAN faces several challenges, in particular to manage the slices and associated control and data planes and for scheduling and resources allocation mechanisms. MAESTRO-5G project develops enablers for implementing and managing slices in the 5G radio access network, not only for the purpose of serving heterogeneous services, but also for dynamic sharing of infrastructure between operators. For this aim the project puts together exerts on performance evaluation, queuing theory, network economy, game theory and operations research. MAESTRO-5G is expected to provide: •A resource allocation framework for slices, integrating heterogeneous QoS requirements and spanning on multiple resources including radio, backhauling/fronthauling and processing resources in the RAN. •A complete slice management architecture including provisioning and re-optimization modules and their integration with NFV and SDN strata. •A business layer for slicing in 5G, enabling win-win situations between players from the telecommunications industry and the verticals, ensuring that the 5G services are commercially viable and gain acceptance in the market. •A demonstrator showing the practical feasibility as well as integration of the major functions and mechanisms proposed by the project, on a 5G Cloud RAN platform. The enhanced platform is expected to support the different 5G services (eMBB and IoT) and to demonstrate key aspects of slicing, such as: - Ability to create and operate in parallel multiple slices, on the same infrastructure and sharing the same radio resources (e.g. spectrum), each having different service requirements. - Ability to create and operate in parallel and independently different slices, sharing the same infrastructure/spectrum, belonging to different business actors, such as different operators. - Demonstrate inter-slice control ensuring respect of SLAs and a fair resource sharing.

The number of languages spoken in Africa ranges from 1 000 to 2500, according to estimates and definitions. Monolingual States do not really exist on this continent since languages cross borders. The number of languages varies from 2 or 3, in Burundi and Rwanda, to more than 400 in Nigeria. Multilingualism is indeed ubiquitous in sub-Saharan African societies. To support the development and use of languages, many institutions and organizations have been created, often under the auspices of the UNESO or the African Union. In summary, the major issues met by these initiatives are: -the development and standardization of linguistic resources in many languages, not just the higher-resourced ones, -the introduction of national languages in the digital space through the creation and dissemination of content in local languages, -the multilingual access to digital resources. If equipped with linguistic and computer resources, languages having a written form can be integrated into the development products of major players in the digital world, attracted by a market with great economic potential. For instance, the mobile phone manufacturers offer more and more models with textual and graphical interfaces in African language. Nevertheless the use of written / textual interfaces requires to be literate! According to Denis Gikunda, director of the development program in African languages at Google, one of the highlights of the online market development in Africa is to ensure that applications talk to Africans in the true sense of the word. Several publications of UNESCO make explicit reference to the speech synthesis (and recognition) as a technological facilitator (one can cite, for instance, the following: “The illiteracy rate remains high: the use of voice interfaces is relevant”). Thus, today is very favorable to the development of a market for speech in African languages. People's access to ICT is done mainly through mobile (and keyboard) and the need for voice services can be found in all sectors : from higher priority (health, food) to more fun (games, social media). For this, overcoming the language barrier is needed and this is what we propose in this project where two main aspects are involved: fundamentals of speech analysis (language description, phonology, dialectology) and speech technologies (ASR and TTS) for African languages. ALFFA project is really inter-disciplinary since it not only gathers technology experts (LIA, LIG, VOXYGEN) but includes fieldwork linguists / phoneticians (DDL). Such a partnership is very important since we want to reuse the strong experience of field linguisists in data collection, as well as their knowledge on dialectal/regional differences, particularly important in Africa. In the project, developped ASR and TTS technologies would be used to build micro speech services for mobile phones in Africa (for instance, a phone service to consult the “price of commodities” or provide “voice reporting for information systems”). If accepted, the ALFFA project would help a young start-up (Voxygen) to interact with academics on the fundamentals aspects of African languages and start deploying prototypes / services in a continent where the telecom market has a strong potential. In addition, the project would help the academic partners to reach an international leadership in the domain of speech processing and analysis for African languages which will reinforce their (already large) collaboration network on this continent. On this purpose, subcontracting is planned in the framework of the ALFFA project in order to set up sustainable collaborations with local actors (academics, NGO) in Africa. The scientific challenges associated to the ALFFA project are detailed in the ANR project proposal form.

In the NICETWEET project, we develop a new, complete, and unified methodology to address an important and generic problem that appears in economics, finance and politics. Several decision-makers are in competition for propagating ideas or selling goods, services, etc. to a large number of agents who are connected through a physical or digital social network. The agents are thus under the (endogenous) influence of their neighbors in the social network graph but also under the exogenous influence of the decision-makers. The latter have a certain knowledge about the social network and the underlined opinion dynamics and they use it for target advertising purposes. To address this problem, we have formed an interdisciplinary team that possesses the required expertise: control theory and more specifically opinion dynamics, game theory, information theory, complex networks, and economics. In contrast with the closest works, we assume that the social network possesses new features that correspond to the economic applications of interest that are analyzed. It is very important to note that we also assume the presence of multiple decision-makers. Indeed, our preliminary knowledge about economic applications indicates that some key features need to be accounted for in a simultaneous manner. Some of these features are: the social network is often large and sparse; agents enter or leave the network randomly at any time; decision-makers may have imperfect knowledge of the social network and opinion dynamics parameters; some agents may exchange not only their opinion but also their reliability; possible presence of extremists. A salient feature with respect to the state-of-the art is that opinion dynamics over the social network can be controlled and several decision-makers try to control it. To address this problem of controlled opinion dynamics in presence of multiple decision-makers who typically have non-aligned utility functions, we will resort to game theory and contribute to bridging the gap between the formal literature of control and the typically non-formal literature of economics and marketing. More specifically, one of our technical goals is to account for the social network and opinion dynamics mathematical models to construct a formal and systematic way of designing efficient and implementable viral marketing strategies for decision-makers. One important issue in the design of viral marketing strategies is the way of allocating a given advertising, campaign, or influence budget among the social network agents and over time (in presence of competition). Efficiency will be measured in terms of using the available information by a given decision-maker and by its way of reacting to the behavior of the other decision-makers. To design viral marketing strategies in the stochastic (repeated) game framework, several approaches will be adopted in parallel to manage the risk aspect. Considering partial information about the social network and opinion dynamics, one approach will be to extract insights from the derived limiting performance characterization theorems namely, theorems that characterize the achievable long-term utilities under partial information. A step further will be to characterize equilibrium utilities and design practical equilibrium space-time budget scheduling strategies i.e., to specify exactly how a decision-maker should allocate his advertising budget among the social network agents and over time. Another approach will be to exploit multi-player learning techniques such as Bayesian learning rules tailored to the problem of interest. To implement the ambitious road-map of NICETWEET and conduct the corresponding research, recent and promising results obtained within the consortium will be exploited as a foundation of the project.

A major issue of language processing evaluation metrics concerns the fact that they are designed to globally mesure a proposed solution from a considered reference, with the main objective of being able to compare systems with each other. While automatic systems are aimed at end-users, they are ultimately little studied: the impact of these automatic errors on the human, and the way in which they are perceived at the cognitive level, has then never been studied, and ultimately not integrated into the evaluation process. The DIETS proposes to focus on the problematic of diagnosis/evaluation of end-to-end automatic speech recognition (ASR) systems by integrating human reception of transcription errors from a cognitive point-of-view. The challenge is here twofold: 1) to analyze finely ASR errors from a human reception, and 2) to understand and detect how these errors manifest themselves in an end-to-end ASR framework, whose work is inspired by how the human brain works.