Call for applications – Doctoral contracts in Artificial Intelligence

In the framework of the IADoc@UdL program, winner of the call for programs "PhD contracts in Artificial Intelligence" launched by the French National Research Agency (Agence nationale de la recherche – ANR) in 2019, the Université de Lyon (UdL) and the partner institutions will co-finance 7 PhD contracts for 36 months.

A call for applications is open until April 20, 2021 for the award of 7 PhD contracts.

#Neural networks, #photonics #artificial intelligence

Supervisors of the thesis: Fabio Pavanello, Alberto Bosio

Laboratories involved: Institut des Nanotechnologies de Lyon(UMR5270)

Project summary:

Many efforts are currently ongoing to explore and demonstrate novel unconventional (non-Von-Neumann) computing architectures. Specifically, brain-inspired (neuromorphic) hardware architectures can deliver several orders of magnitude superior performance in terms of energy efficiency, computation density, speed and latency compared to classical CPU and GPU-based computing solutions. Several applications ranging from 5G technology and IoT processing to autonomous driving and robotics would greatly benefit from such solutions.

The focus of this thesis will be to develop neuromorphic architectures leveraging emerging technologies suitable for the implementation of these novel computing paradigms.

Among the myriad of technologies under investigation, integrated photonics is regarded as one of the best candidates because of its large potential in terms of parallelization, high-speed operation and speed-of-light propagation as well as low power consumption and large number of physical degrees to manipulate/encode the information (amplitude, polarization, phase, etc.). Besides, integrated photonics benefits from CMOS-compatible fabrication for volume scaling and ease of market take-up.

In the above context, this PhD thesis aims to explore the limitations in terms of robustness and reliability of photonic neural networks.

Detailed description of the thesis

#Machine learning #optimal transport #enhancer #gene expression #embryonic development

Supervisors of the thesis: Yad GHAVI-HELM, Paul VILLOUTREIX, Cédric VAILLANT

Laboratories involved: Institute of Functional Genomics of Lyon (UMR 5242), Physics Laboratory of ENS de Lyon (LPENSL) ( (UMR5672), Laboratoire d'informatique et systèmes (LIS) (UMR7020) à Marseille

Project summary:

In recent years, machine learning algorithms have been increasingly applied to data-driven sciences such as genomics, and have proven particularly relevant for making sense of the wealth of data generated by omics techniques in the field of developmental biology.

In this interdisciplinary project, we propose to use an unsupervised learning method to study the regulation of gene expression during Drosophila embryogenesis.

More specifically, we will adapt a method based on the mathematical framework of optimal transport to reconstruct the spatio-temporal activity of enhancers.We will then develop an interactive 3D atlas of enhancer activity and gene expression to visualize the results and use this data to further improve the prediction of enhancers’ target gene(s).

Detailed description of the thesis
 

#Reinforcement learning #6G #ML-based wireless networks

Supervisors of the thesis: Jean-Marie GORCE, Malcolm EGAN

Laboratories involved: Centre of Innovation in Telecommunications and Integration of service (CITI) (EA 3720)

Project summary:

The association of machine learning (ML) and wireless communications is a topic that emerged in early 2000, under the cognitive radio paradigm. However, recent progress in deep learning, reinforcement learning and federated learning has opened up fascinating new perspectives for wireless systems.

Massive access for machine type communication is a typical setup where the fundamental constraints such as latency, reliability or rates are too complex to guarantee under reasonable bandwidth and energy resources. The classical approach to optimize radio communications involves designing a multi-layer protocol that satisfies tradeoffs between the different constraints, based on services requirements.

However, these requirements have evolved, making the protocols out-of-date. Designing massive access protocols able to self-adapt to the evolution of the constraints is one of the most important challenge for 6G. This PhD will explore algorithmic solutions to the massive access problem in order to develop data-driven scheduling and resource allocation algorithms, able to account for complex dependencies in data traffic and time varying network statistics.

Detailed description of the thesis

#Ordinal data #categorical data #textual data #model based clustering #dynamic model

Supervisors of the thesis: Julien JACQUES, Isabelle PRIM-ALLAZ

Laboratories involved: Laboratoire ERIC (UR 3083), Laboratoire COACTIS (UR4161)

Project summary:

In many areas of humanities and social sciences the studies are based on questionnaires completed by participants, often repeated along time. The data provided by the answer of participants are of different nature, generally quoted as mixed data in the literature (nominal or ordinal categorical data, quantitative data, textual data). In addition, these mixed data here have a temporal component that it will be necessary to take into account.

The goal of the PhD thesis is to propose a new unsupervised machine learning model (clustering) which is able to extract typical behaviours from the temporal mixed data sets. Thus, the data analysis will no longer be based on the observation of the individual responses to the questionnaires, but on the summaries provided by the clusters which gather set of participants which have the same evolution of answers over time. This information is essential for data analysis from a humanities and social sciences point of view.

If some works already exist for analysing mixed data set, especially in a probabilistic modelling framework, two main scientific challenges remain: "how to manage the weight of the different type of data?" "How to model the time evolution?" The goal of the PhD thesis is to provide solutions to these two challenges. 

Detailed description of the thesis

#federated machine learning #healthcare #privacy and security #medical imaging

Supervisors of the thesis: Stefan DUFFNER, Carole LARTIZIEN

Laboratories involved: Centre de Recherche en Acquisition et Traitement de l'Image pour la Santé (CREATIS) (UMR 5220 U1206), Laboratoire d'Informatique en Images et Systèmes d'Information (LIRIS)

Project summary:

Application of machine learning (ML) to healthcare is among the most challenging ones with the potential to exploit information provided by an exponentially growing mass of heterogeneous data (images, semantic information, biological parameters,..). Those models require a large amount of data to perform well, particularly in the era of large-scale deep neural networks. One option to increase the training population is to promote multi-centre clinical studies, which opens many privacy-related problems since data producers lose control over their data as well as huge data traffic.

Federated learning (FL) is a new ML approach that was recently introduced to counterbalance the need to access large databases by the responsibility to maintain the privacy of individual participants. In this context, FL appears as a very promising technique, first to account for patient privacy thus complying with the increasingly stringent general data protection regulations (GDPR) and then to limit the huge amount of data traffic required when gathering medical data to a centralized server.

This research field is in its early premise and needs to address key challenges related to the specificity of medical data. The aim of this PhD is to investigate methodological research in this domain with application to the design of diagnosis and prognosis models of brain pathology based on multimodality imaging.
Detailed description of the thesis

#data integration #data analysis #data driven prediction #air pollution #respiratory health #hospital admissions

Supervisors of the thesis: Delphine MAUCORT-BOULCH, Mohand Said HACID

Laboratories involved: Laboratoire de biométrie et biologie évolutive (LBBE) (UMR 5558), Laboratoire d'Informatique en Images et Systèmes d'Information (LIRIS)

Project summary:

Air quality is a major issue for health and environment. Its impact is estimated at more than 50,000 premature deaths per year and the cost of air pollution is between 70 and 100 billion euros per year. Faced with such a situation, the concerned stakeholders must combine their efforts to develop approaches/tools and know-how to:

1. better identify and characterize pollutants and their sources,
2. predict their impact on the concerned services (e.g., hospitals). 

This thesis falls within this context. The ultimate goal would be to derive a model for predicting emergency and intensive care admissions as a function of peak concentrations of particulate matter and atmospheric conditions. This objective therefore targets the issue (2). It consists in understanding the problem of air quality prediction and its impact on the management of hospital admissions, in particular, at the level of hospital emergencies, intensive care units, and well as pulmonary and cardiovascular diseases.

We will develop innovative methods and tools allowing a continuous and real-time assessment of air quality (in urban areas) and its impact on emergency room admissions for a better organization and management of hospital resources - the results of this research could be used to better predict hospitalization patterns and costs for the health system and to trigger an increased vigilance on particle pollution in cities.

Detailed description of the thesis

#Deep neural networks #game theory #optimization

Supervisors of the thesis: Marc SEBBAN, Charlotte LACLAU, Ievgen REDKO

Laboratories involved: Laboratoire Hubert Curien (UMR CNRS 5516)

Project summary:

The theoretical analysis of deep neural networks (DNN) is arguably among the most challenging research directions in machine learning right now, as it requires scientists to lay novel statistical learning foundations to explain their behavior in practice. In this proposal, we aim to explore the interplay between DNNs and game theory by considering the widely studied class of congestion games with thegoal of relating them to both linear and non-linear DNNs and to the properties of their loss surface. Beyond retrieving the state-of-the-art results from the literature in a principally new way, we expectthat our proposal will provide a very promising novel tool for analyzing DNNs and will allow solving concrete open problems related to:

1. characterizing the DNNs optimization inefficiency depending onthe algorithmic choices, such as their architecture, activation, and loss function used,
2. proposing new optimization strategies with strong convergence guarantees.

Detailed description of the thesis