The defense will take place on wednesday 17th of september.

Title
Hardware Arithmetic Acceleration for Machine Learning and Scientic Computing.

Abstract
In a data-driven world, machine learning and scientific computing have become increasingly important, justifying dedicated hardware accelerators. This thesis explores the design and implementation of arithmetic units for such accelerators in Kalray’s Massively Parallel Processor Array.

Machine learning requires matrix multiplications that operate on very small number formats. In this context, this thesis studies the implementation of mixed-precision dot-product-and-add for various 8-bit and 16-bit formats (FP8, INT8, Posit8, FP16, BF16), using variants of a classic state-of-the-art technique, the long accumulator. It also introduces techniques to combine various input formats. Radically different methods are studied to scale to the larger range of 32-bit and 64-bit formats common in scientific computing.

This thesis also studies the evaluation of some elementary functions. An operator for exponential function (crucial for softmax computations) extends a state-of-the-art architecture to accept multiple input formats. The inverse square root function (used for layer normalisation) is accelerated by combining state-of-the-art techniques for range reduction, correctly rounded multipartite tables, and software iterative refinement techniques.

Date: 24/07/2025
Time: 10 am
Room: Amphitheatre Claude Chappe
Venue: Batiment Hedy Lamarr

Title
A study of energy consumption challenges in extended reality services over cellular networks

Abstract
Extended reality (XR) services are characterised by their heavy computational requirements throughout their life cycle. XR comprises multiple traffic modes consisting of 3D video and audio, haptics, sensor and pose information. Systems-related challenges relating to the creation, encoding, transmission, rendering and presentation of the application data from this class of services have increasingly become key areas of research inquiry from both a computational and energy viewpoint. Recently, new provisions have been made across different relevant standards to improve the capacity of these applications on mobile cellular networks. Our research investigates the energy-related challenges at both the RAN and UE levels.  We conduct our evaluations using system-level simulations (SLS) that adhere to the parameter specifications established by standardisation bodies.
We begin with a RAN-level energy reduction plus XR application enhancement strategy. With the cell switch off (CSO) technique proposed for BS deployment energy efficiency at low load, we evaluate the impact of an XR-capacity improvement criterion on the energy savings obtainable. Our analysis covers a European urban city using data from a European operator and system-level simulations according to the standard network operations.
In XR, UE energy-saving schemes are especially important since current delivery devices are still in early commercial development, making battery-saving techniques crucial. In this thesis, we use the Rel-18 improvements on the discontinuous reception (DRX) UE energy saving mechanism through state-of-the-art prediction algorithms to propose a traffic prediction-based non-integer DRX mechanism. We achieved significant energy savings without impeding the XR quality of service requirements.
Lastly, towards enhancing the capacity of XR on cellular networks in energy-saving mode, we propose a DRX plus QoE-aware scheduler (DQAS). Using SLS according to 3GPP specifications on XR QoE requirements, traffic model, and dense urban deployment scenarios, and following real traces from XR applications, we first evaluate a QoE-aware scheduler (QAS) for XR services. We introduced DRX awareness into QAS, jointly improving XR QoE and energy consumption. We analysed our results to identify the parameter window in which both metrics can be improved towards achieving the goal of XR capacity improvement on cellular networks.

Jury
– Riadh Dhaou, Professor des Universites (University of Toulouse), Rapporteur
– Thi-Mai-Trang Nguyen, Professeur des Universités (Sorbonne Paris Nord University), Rapporteure
– Nadjib Achir, Maître de Conférences HDR(Sorbonne Paris Nord University), Examinateur
– Herve Rivano, Professeur des Universités (INSA Lyon), Examinateur
– Razvan Stanica, Maître de Conférences HDR (INSA Lyon), Directeur de thèse

When
Monday 30 June at 10:00 am

Where
AE1 amphitheater, Gustave Ferrié building, 8 rue de la Physique, F-69621 Villeurbanne

https://insa-lyon-fr.zoom.us/j/95728542338

Title
Design of radio front-end architectures to meet low energy consumption or resource sharing objectives

Abstract
The contributions presented are structured around four main research areas :
i) the energy consumption of connected devices, with the proposal of asynchronous wake-up radio solutions;
ii) the harvesting of ambient electromagnetic energy;
iii) the design of UHF RFID tags that are robust to elongation or integrate sensing capabilities (humidity/temperature), and on the other hand, studies on the tag-to-tag communication concept;
iv) the reduction of the dynamic range of input signals of analog-to-digital converters, as well as the joint implementation of spatial modulation and a full-duplex mode.
Depending on the addressed topic, the studies presented in this manuscript include theoretical, simulation, experimental, and real-world testing, with particular attention to the non-idealities of the radio front-end building
blocks and their impact on performance.
Based on the acquired expertise and the collaborations conducted at the national and international levels in various contexts, the final section identifies several new research directions that are being considered.

Jury
Prof. Daniela DRAGOMIRESCU INSA Toulouse ; reviewer
Prof. Ioannis KRIKIDIS Université de Chypre ; reviewer
Prof. Smail TEDJINI Université Grenoble-Alpes, Grenoble INP; reviewer
Prof. Yvan DUROC Université Claude Bernard Lyon 1 ; examiner
Prof. Bruno ALLARD INSA Lyon ; examiner
Prof. Guillaume VILLEMAUD INSA Lyon ; examiner

When
The defense will take place on june 6th at 9 AM.

Where
INSA Lyon Campus,  CITI Laboratory, Amphi Chappe,  Building Hedy Lamarr, 6 Avenue des Arts 69621 Villeurbanne, France

https://insa-lyon-fr.zoom.us/j/91634868780?pwd=i5znUHEaIiuahaoIkrkL0g5DU9fEaD.1

Title
MultiScale-HPC Hybrid Architectures: Developing Computing Continuum Towards Sustainable Advanced Computing.

Jury
Prof. Eddy CARON (UCBL, Lyon, France)
Prof. Ewa DEELMAN (USC, Los Angeles, United States of America)
Prof. Michaël KRAJECKI (URCA, Reims, France)
Prof. Genoveva VARGAS SOLAR (CNRS, Lyon, France)
Prof. Bernd MOHR (JSU, Jülich, Germany)
Prof. Frédéric DESPREZ (INRIA, Grenoble, France)
Prof. Frédéric LE MOUËL (INSA, Lyon, France)

The defense will take place on mai 19th at 5 PM.

Title
Space Networking: Models and Methods

Where
INSA Lyon Campus, Amphi Chappe

The defense will take place on wednesday 5th february at 2 PM in the Heidi Lamarr building (Amphi Chappe), Insa-Lyon, Villeurbanne.

Title 
Seamless Continuous Integration / Continuous Delivery (CI/CD) for Software Defined Vehicles

Abstract
Driven by the rapid increase in the number of Electronic Control Units (ECUs), current automotive software systems face growing complexity while advancements in software architecture are well behind. This imbalance has resulted in higher system complexity, important financial costs, and significant challenges in maintaining and deploying new services in vehicles. The thesis explores the potential of adopting Continuous Integration/Continuous Delivery (CI/CD) pipelines for software-defined vehicles, focusing on several critical aspects: secure software deployment, adaptability of in-vehicle software, and optimization of performance using edge computing.

The contributions of the thesis are manifold: (1) A comprehensive taxonomy of key findings related to the transformation of automotive ICT systems, (2) A proposal for a blockchain-based multi-automaker software store to manage updates and dependencies, (3) The development of a virtualization framework for multi-microcontroller systems and an evaluation of these OS-level virtualization solutions for in-vehicle systems, (4) A software orchestration framework that prioritizes criticality and optimizes resource allocation in heterogeneous environments, and finally (5) A consensus algorithm to efficiently offload functions to edge-computing IoT nodes, optimizing resource use in automotive cloud-edge systems.

By addressing these issues, the thesis contributes to the future of automotive ICT systems, proposing innovative methods that strike a balance between flexibility and performance in managing software complexity within the evolving landscape of connected, autonomous vehicles.

Jury

• Sara BOUCHENAK, Professeure des Universités à L’INSA de Lyon
• Diala NABOULSI, Professeure Associée à l’ETS Montréal, rapporteuse
• Thierry DELOT, Professeur des Universités à l’Université Polytechnique Hauts-de-France, rapporteur
• Hadi TABATABAEE, Professeur Assistant à l’UCD Dublin Jean-Marc MENAUD
• Professeur des Universit ́es à l’IMT Atlantique
• Frédéric LE MOUËL, Professeur des Universités à l’INSA de Lyon, Directeur de Thèse
• Tista LIN, Architecte Logicielle à STELLANTIS, Co-encadrante de thèse

The defense will hold on monday, December 16th, at 9.30 AM in room ,020G at ENSSAT Lannion

Title
Powering low-power Wake-up Radios with RF energy harvesting.

Abstract
Due to the massive deployment of connected devices in the context of the Internet of Things (IoT), powering them exclusively with cables or batteries is not efficient. This thesis explores the use of radiofrequency (RF) energy as an alternative power source for wake-up radios (WuRx) in wireless sensors, thereby reducing their reliance on batteries. The first challenge is to develop an RF energy harvesting circuit capable of providing a regulated voltage from low power levels. An innovative solution is proposed, based on Schottky diode RF rectifiers incorporating the inductive technique. This circuit ensures the operation of an energy management system that powers a semi-active WuRx and stores excess energy when higher power levels are available.

Given the intermittent nature of RF energy, the second challenge is to adapt the WuRx’s energy consumption by modulating its quality of service, defined as the percentage of processed signals among those received, based on the harvested energy.

Jury
* Nathalie DELTIMPLE, Professor at Bordeaux INP, Reviewer
* Christian VOLLAIRE, Professor at Ecole Centrale Lyon, Reviewer
* Daniela DRAGOMIRESCU, Professor at INSA de Toulouse, Examiner
* Laurent CLAVIER, Professor at IMT Nord Europe, Examiner
* Dominique MORCHE, Research Director at CEA-LETI, Invited
* Matthieu GAUTIER, Professor at Univ. de Rennes,Thesis Director
* Guillaume VILLEMAUD, Assoc. Prof. at INSA de Lyon, INSA de Lyon,Thesis Co-Director
* Olivier BERDER, Professor at Univ. de Rennes,Supervisor
* Florin-Doru HUTU, Assoc. Prof. at INSA de Lyon, INSA de Lyon,Supervisor

The defense will take place on Monday December 16 at 14h in the Amphi Huma Ouest at Insa Lyon.

Title
Uplink Resource Allocation Methods for Next-Generation Wireless Networks

Abstract 
Facing the diversity of communication needs of 5G networks and the future 6G, resource allocation is considered as a key enabler to increase the number of devices, the data rate or the reliability of the communication links. In MTC networks, recent work has proposed to adapt the temporal resource allocation as a function of the underlying process
driving the activity of the devices. This thesis firstly focuses on the impact of having only limited knowledge of the underlying process, and proposes methods to mitigate the bias induced by the lack of knowledge.
Secondly, an algorithm for the joint optimization of the temporal resource allocation and the transmit power of the devices is proposed. The algorithm ensures that devices that are likely to transmit on the same resources do so with a sufficient power diversity to ensure their decodability by the BS. Finally, in networks with an eMBB objective, we
propose to jointly optimize the power, the frequency resources used, as well as the number of parallel data streams used by the devices. Our simulation study shows that our joint optimization outperforms current 5G baselines for which these parameters are common to all devices of the cell.

Jury
* LOSCRI Valeria, Directrice de Recherche, Inria Lille, Rapporteur
* LIVA Gianluigi, Chercheur, German Aerospace Center, Rapporteur
* POPOVSKI Petar, Professeur, Aalborg University, Examinateur
* FIJALKOW Inbar, Professeure, ENSEA, Examinatrice
* VALCARCE Alvaro, Ingénieur de Recherche, Nokia Bell Labs, Examinateur
* ADJHI Cédric, Chargé de Recherche, Inria Saclay, Examinateur
* GORCE Jean-Marie, Professeur, INSA Lyon, Directeur de thèse
* EGAN Malcolm, Chargé de Recherche, Inria Lyon, Co-encadrant

The defense will take place the 5th december at 9 AM at the library Marie-Curie INSA-Lyon

Title
Health Data: Exploring Emerging Privacy Enhancing Mechanisms

Abstract
Health data represents a large volume of information, generated daily and sensitive by nature. However, sharing this data is essential for advancing research and, ultimately, improving patient care. The use of medical data faces limitations due to its sensitivity and the need to ensure confidentiality, which is governed by current regulations. This
necessitates enhanced protection. Interest in alternatives to sharing raw data, such as pseudonymization or anonymization, is increasing alongside the growing need for access to training data for the use of artificial intelligence, which requires large amounts of data to function effectively as a medical assistant.

In this thesis, we explore new privacy-preserving mechanism made possible by the rapid advancements in artificial intelligence. More specifically, my analysis focuses on improving alternatives to the centralization of sensitive data: federated learning, a decentralized method of training artificial intelligence models that do not need sensitive data sharing, as well as synthetic data generation, which creates artificial data similar statistical properties to real data.
Given the lack of consensus on evaluating the privacy of these new approaches, our work focuses on the systematic measurement of privacy leakage and the balance with the utility of synthetic data or the federated learning model. My contributions include a mechanism to enhance the privacy properties of federated learning, as well as a new method for conditional synthetic data generation. This thesis aims to contribute to the development of more robust frameworks for the secure sharing of health data, in compliance with regulatory requirements, thereby facilitating innovations in healthcare.

Jury
* Sonia BEN MOKHTAR, Directrice de Recherche, CNRS/INSA-Lyon, Examiner,
*Szilvia LESTYAN, Docteure-Ingénieure de Recherche, INRIA, Examiner,
* Jérémie DECOUCHANT, Professeur des universités, Université de Delft, Examiner,
* Benjamin NGUYEN, Professeur des universités, INSA-CVL,Thesis Reviewer,
* Emmanuel VINCENT, Directeur de Recherche, INRIA,Thesis Reviewer,

The defense will take place on december 10th at 1:30 PM.

Title
Privacy issues in AI and geolocation: from data protection to user awareness

Abstract
The evolution of digital technologies and their increasing adoption have opened major opportunities, highly beneficial for society in general and for individuals in particular. However, it also poses considerable threats to privacy that require appropriate legal and ethical rules. Privacy is essential to protect individuals, for example against possible misuse of personal data. Privacy is also essential to protect society, as shown by the misuse of personal data to influence voters
during elections (e.g., Cambridge Analytica).
In this context of ultra-rapid development of technologies (often deployed before being regulated), my research work is focused on privacy protection. More precisely, I mainly contribute to the field by proposing technical solutions to privacy (by quantifying risks or proposing countermeasures for example), and also through transdisciplinary activities. Indeed, privacy issues cannot be solved by technology alone because they also raise legal, ethical, economic and societal questions that require a dialogue with people from different disciplines.
My main contributions cover 1) issues related to the collection, exploitation and protection of location data, and more recently 2) security and confidentiality of AI. In this second axis, I focused on “privacy considerations in ML”, i.e., the identification of risks related to ML technologies and countermeasures, and “exploiting ML for confidentiality”, using the capabilities of these new tools to protect individuals (with the use of language models for the anonymization of
medical reports for example).
To address these growing privacy issues, it is necessary to quantify the new risks fueled by new technologies and new usages, and to improve the safeguarding of users’ personal information by developing protection mechanisms. Finally, it is also necessary to both raise awareness among end users about the different risks in order to enable them to adapt
their use, and to collaborate with key players in the field to adopt best practices.

Jury
* Pr. Anne-Marie Kermarrec, EPFL
* Pr. Romain Rouvoy, Université de Lille (rapporteur)
* Dr. Aurélien Bellet, Inria (rapporteur)
* Dr. Catusci Palamidessi, Inria (rapporteuse)
* Pr. François Taiani, Université de Rennes 1
* Pr. Sébastien Monnet, Université Savoie Mont-Blanc
* Pr. Eddy Caron, Université Lyon 1
* Dr. Sonia Ben Mokhtar, CNRS, Insa-Lyon