The aim of this project is to explore the first avenues of research into the contribution of multimodality in datasets to meet the requirements of fair learning. Fairness refers here to the biases (in the data and/or induced), while being interested in the interpretability of the models to help their certification. Each modality has its own statistical and topological characteristics, which requires upstream research on the adjustment of distributions when biased, adapted metrics, etc. Moreover, each one being itself a bias of observation of the data, this will be taken into account to establish a joint distribution (trans-modal) unbiased on all these modalities. With theoretical research in cross-modal statistical learning, we will study methods for reducing some types of identified biases (non iid, imbalances, sensitive variables) in the case of multimodal data. Two levels of treatment are privileged: (1) cross-modal pre-processing of biases in the data, by learning metrics, neural representations, and optimization constraints on kernel pre-images; (2) cross-modal algorithms for eliminating biases in model learning: cross-modal optimization algorithms, as well as optimal transfer and transport approaches between modalities to debias the concerned ones, based on the theoretical results previously obtained. Parsimony will be considered for scaling and explainability. Transversally, our work will be based on problems arising from real data sets in biology and health, multi-modal and presenting various types of bias, and on toy data sets to be generated. They have modalities where the data are structured in graphs: all our fundamental works will be declined to take into account this specificity impacting the treatment of the considered biases.

The Internet of Things connects physical devices offering sensing or actuating with their vicinity. The ever-growing capabilities of devices allow to imagine new architectures including them as first class citizens. New added-value applications can then be envisioned in smart agriculture, smart buildings, smart cities, energy and water management, e-health and ageing well... The Web of Things (WoT) allows to describe the devices semantics, bridging the gap between the different domain and service descriptions. In today WoT architectures, physical devices can be located at distance from systems that perform reasoning. A centralised approach does not take advantage of the devices capabilities and induces suboptimal data transfers as well as server overload. Besides, many devices are now smart enough to discover each other, exchange data, and collectively make decisions. CoSWoT objectives are to propose a distributed WoT-enabled software architecture embedded on constrained devices with two main characteristics: (1) it will use ontologies to specify declaratively the application logic of devices and the semantics of the exchanged messages; (2) it will add reasoning functionalities to devices, so as to distribute processing tasks among them. Doing so, the development of applications including devices of the WoT will be highly simplified: our platform will enable the development and execution of intelligent and decentralised smart WoT applications despite the heterogeneity of devices. In CoSWoT, WoT applications will rely on a platform hosting the base services. Besides traditional services, it will host extensions that correspond to two scientific barriers: (1) the use of ontologies as a generalised model for exchanges between heterogeneous devices. A joint statement from AIOTI WG3, IEEE P2413, oneM2M, W3C positions ontologies as key enablers for semantic interoperability on the WoT. However research questions remain concerning (i) the adequation of existing ontologies to the target application domains; (ii) the applicability of theoretical principles developed in a variety of protocols and standards, in the context of data streams; (iii) the discovery of heterogeneous devices, their services and how to solicit them. (2) distributed and embedded incremental reasoning. Devices become powerful enough to offer storage and processing; new architectures appear, based on edge computing including devices such as sensors and actuators. The data streams provided by sensors require to perform incremental reasoning tasks. Research questions remain on (i) how to embed reasoning in devices with various capacities, it requires specific optimisations; (ii) how to efficiently distribute reasoning tasks among devices. Smart agriculture is a typical application domain of such WoT architectures, where the surveillance of cultivated fields requires various sensors that push streaming data, which must be collected and reasoned upon to take decisions executed by actuators. Smart buildings is another such typical application domain where added-value application services involve other verticals such as energy management, e-health, or ageing well. We will define use cases and requirements for smart agriculture and smart buildings, run simulations, and then lead real experiments. The CoSWoT platform will foster the decoupling of the development of software and the development of hardware, so as to ease the emergence of a new economic sector in the digital industry around WoT applications development, disconnected from the development of the smart devices themselves.

Dry eye syndrome (DES) is a worldwide disease that affects the quality of life of millions of individuals to varying degrees. If the positive diagnosis is generally very easy, the objective and reliable determination of its severity (grading) remains the main diagnostic problem for decades. In practice, two very different situations must be distinguished: routine clinical practice where grading is necessary to personalize treatment and clinical trials where grading is essential to demonstrate the efficacy of a new treatment: 1/ In routine clinical practice, the clinician collects the symptoms, observes the ocular surface stained with fluorescein to assess superficial punctate keratitis and tear film break-up time and estimates the quantity of tear using the Schirmer test. 2/ During clinical trials, the very high imprecision and/or the lack of reliability (precision, accuracy, repeatability) of the 3 tests forces the addition of other evaluation criteria, like different biomarkers but none of which has ever been imposed or transferred to routine use. FLUOSICCA aims to revolutionize the diagnosis of the severity and monitoring of dry eye syndrome by developing a process combining fluorescence imaging and biology. We will quantify in situ a pair of biomarkers, directly on the ocular surface of patients in a sensitive and specific manner: a marker of constant or little variation in the pathology and serves to normalize the signal and a marker that is overexpressed in dry eye syndrome. The measurement of a ratio will reduce inter and intra individual variability. The ligands of the two biomarkers (Partner 3) will be synthetic, non-immunogenic, stable and easily customizable proteins. They will be coupled to functionalized BODIPY fluorophores (Partner 2), excitable in a non-dazzling wavelength range (>650 nm, easy to tolerate) while maintaining brightness and photostability. The imaging device (Partner 4) will be sensitive and can be industrialized on a large scale. It will allow the detection and discrimination of the two fluorophores in order to calculate their ratio. The whole system will first be developed in vitro on cells expressing the two biomarkers, then tested on primary ocular epithelial cells and on human corneas preserved in partner 1's bioreactor, using an innovative dry eye ex vivo model. This integrated solution will then be transposable to other biomarkers and other pathologies of the ocular surface.

Security failure in computing systems has become one of today’s biggest concern. The primary threat is the fact that modern computing architectures –from computational optimizations to storage elements and interfaces, from end-user applications to the operating system & hypervisor, and from microarchitecture to underlying hardware– may hide unexpected vulnerabilities. This concern is gaining further momentum, with the spectacular aggressiveness of Spectre, Meltdown and ZombieLoad vulnerabilities. They demonstrate that even hardware, which is often considered as an abstract layer that behaves correctly by executing instructions and giving a logically correct output, is leaking critical information as a side effect of software implementation and execution. Even worse, the many undocumented parts of modern architectures open doors for yet undescribed side channel attacks. This proposal tackles the problem of these vulnerabilities at the intersection of software and hardware to propose a secure-by-design computing where we strike a balance between security and hard earned performance benefits.

The GeSPAD project is devoted to achieve a major breakthrough in the development of the next-generation of Single Photon Avalanche Diodes (SPAD), a high-sensitivity optoelectronic detector with many applications in X-ray tomography, biochemical sensors, DNA/protein microarray scanning, engine/turbine design, aids for disabled people, high-sensitivity and low-light cameras. In the near future, arrays of low-area SPADs will be employed for new applications in the Internet-of-Things such as user detection in smart wearable devices or in the optimization of long distance ranging for autonomous car applications. Today’s market is dominated by silicon-based SPADs, whose sensitivity is limited to photon wavelengths lower than 1100 nm. However, in order to improve depth accuracy in LiDAR systems it is highly desirable to shift the operation wavelength from 900 nm to 1500 nm, thus allowing the use of higher laser powers in compliance with eye-safety specifications. From an industrial perspective, the germanium option is a promising solution for integration on conventional CMOS process, but significant efforts in terms of technology development are needed, and a clear benchmark of the Ge-SPAD performances against III-V and Si-based devices is still missing. The main objective of GeSPAD is to assess novel designs of germanium-based SPAD devices, matrices and circuits and to benchmark them with their III-V and silicon counterparts. The devices designed within this project will feature enhanced infra-red photon detection probability (high quantum efficiency for wavelengths around 1310-1500 nm) as well as low dark count rates, noise and jitter. To attain this objective, Ge-based SPADs will be inspected at all possible levels, going from material properties through device physics until circuit optimization. This project will combine advanced characterization techniques on industrial Ge-based prototypes with multi-scale predictive simulation tools, and efficient quenching circuit design. The consortium will be composed by three academic laboratories and an industrial partner, STMicroelectronics, which is one of the major CMOS-SPAD producer in the world, and intends to maintain its leadership by investing in disruptive technologies. The partners are expert in different skills and will adopt complementary methodologies: C2N (coordinator) will study the device physics of SPADs by addressing full-band quantum transport simulations and time-dependent simulations with the particle Monte-Carlo method; LAAS will perform ab-initio calculations on defect properties in Ge and Si/Ge heterostructures as well as their photoluminescence spectroscopy; Lab. Hubert Curien will develop spice models for devices and circuits; STMicroelectronics will provide the electrical characterization of in-house devices as well as TCAD simulations to optimize the SPAD architecture. GeSPAD will have multiple repercussion on the community. From an industrial perspective, it will contribute to the design of the next generation of SPAD architecture and will very probably result in several patents. From an academic perspective, the clarification of the scientific problems here addressed will permit to gain a deeper understanding of the physics of optoelectronic devices and will result in publications in international journals and conferences. Finally, it will facilitate the creation of a French community working around SPADs.