Research projects
On this page, you will discover the research projects underway at Avignon Université, whether national, European or international. If you need assistance in submitting your own current research projects, don't hesitate to contact Projects, Partnerships and International Development Division (DARI) for more information and support.
Current research projects
Programme
ANR - Human and social sciences
Sub-programme
Culture, creativity and heritage
Type of project
Collaborative research project
Duration
4 years
Scientific Manager
Boris Deschanel
Project coordination
Laboratory
UMR 8562 CNE - Dynamics of Social Worlds
Partners
Institutions and historical dynamics of the economy and society - University of Erfurt - Centre for the Modern and Contemporary Mediterranean - Research Unit on History, Languages, Literatures and Interculturalism - English-speaking critical transfers - Centre for interdisciplinary research in social sciences - University of Montreal - University of Leiden
Summary
The use of shares to finance businesses is a feature of the modern era, affecting entire sectors of the economy. The first joint-stock companies were inspired in early 17th-century France by models from England and the United Provinces. The ACTIMOD project is a social history project which, thanks to a database of identified shareholders taking part in the various companies, should make it possible to see the social and cultural characteristics of shareholding in France and its evolution, and to highlight a group of active investors who weigh on the economic and political choices of the State and herald the groups of influence of the contemporary era. They are one of the keys to understanding the economic changes taking place at the end of the modern period.
Reference
ANR-20-CE26-009
Website
Programme
ANR - Digital sciences
Sub-programme
Artificial intelligence
Type of project
Company collaborative research project
Duration
4 years
Scientific Manager
Mohamed MORCHID
Project coordination
Avignon University
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Partners
Institutions and historical dynamics of the economy and society - University of Erfurt - Centre for the Modern and Contemporary Mediterranean - Research Unit on History, Languages, Literatures and Interculturalism - English-speaking critical transfers - Centre for interdisciplinary research in social sciences - University of Montreal - University of Leiden
Summary
The use of shares to finance businesses is a feature of the modern era, affecting entire sectors of the economy. The first joint-stock companies were inspired in early 17th-century France by models from England and the United Provinces. The ACTIMOD project is a social history project which, thanks to a database of identified shareholders taking part in the various companies, should make it possible to see the social and cultural characteristics of shareholding in France and its evolution, and to highlight a group of active investors who weigh on the economic and political choices of the State and herald the groups of influence of the contemporary era. They are one of the keys to understanding the economic changes taking place at the end of the modern period.
Reference
ANR-20-CE26-009
Website
Programme
ANR - Specific
Sub-programme
ERANET CHIST ERA
Type of project
ERANET
Duration
3 years
Scientific manager
Juan-Emmanuel TORRES
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
University of Lorraine
Partners
University of Science and Technology - Deusto
Summary
With the development of information in different media such as television programmes and the internet, a new question arises.
a new question arises. How can a user access information expressed in a foreign language?
foreign language?
The idea behind the project is to develop a multilingual comprehension assistance system without any
the intervention of a human being. What we'd like to do is help people understand
broadcast news presented in a foreign language and compare it with the news available in the mother tongue of the person concerned.
available in the user's mother tongue. The concept of comprehension is addressed
in this project by providing access to all information regardless of the language in which it is presented.
in which it is presented. With the development of the internet and satellite television, tens of thousands of programmes and news programmes are available in different languages.
satellite television, tens of thousands of programmes and news items are available in different languages, it turns out that even people with a high level of education don't speak more than three languages.
It turns out that even highly educated people speak no more than two or three languages, while the majority speak only one.
the majority speak only one language, which makes this enormous amount of information inaccessible. As a result, the
the majority of television and radio programmes and information on the internet are
majority of people. Yet we would like to listen to the news in our own language and compare it to what has been said on the air.
compare with what has been said on the same subject in another language. For example, how do you
the subject of AIDS is presented in Saudi Arabia and the United States? What is the opinion of
Jerusalem-Post on Yasser-Arafat? And how is he presented in Al-Quds? Go to
We are proposing to develop AMIS (Access to Information and Communication).
AMIS will provide a different perspective on the event.
The comprehension process is considered here as understanding the main ideas of a video.
The best way to do this is to summarise the video so that you have access to the essential information.
essential information. From now on, AMIS will concentrate on the most relevant information, summarising it and translating it for the user if necessary. Another aspect of AMIS is to
compare two summaries produced by this system, in two languages, on the same subject
whatever their medium: video, audio or text, and to explain the difference between their
content in terms of information, feelings, opinions, etc. In addition, the coverage of
web and social media will be exploited to strengthen or weaken the
opinions. AMIS could be integrated into a television remote control or software associated with any Internet browser.
In conclusion, AMIS will address the following research issues: Text, audio and video summarisation - Automatic speech recognition (ASR) - Machine translation - Multilingual sentiment analysis - Achieving synergy between the above research themes.
Reference
ANR-15-CHR2-001-04
Website
Programme
ANR- Environmental Sciences
Sub-programme
Food and food systems
Type of project
Young researchers
Duration
4 years
Scientific Manager
Alice CHÂTEAU-HUOT
Laboratory
UPRI
Project coordination
Avignon University
Partners
French National Research Institute for Agriculture, Food and the Environment (INRAE)
Summary
The objectives of the BCLYSIN project are I/ to identify and characterise the genes involved in endolysin ligand biosynthesis, leading to new insights into the general scheme of the complex biochemical pathways that allow cell envelope assembly in B. cereus and the identification of endolysin ligands and II/ to assess the antimicrobial effect of endolysins on biofilm formation in B. cereus and suppressed.
Reference
ANR-23-CE21-0001-01
Programme
ANR-Environmental science
Sub-programme
Living earth
Duration
4 years
Scientific manager
Elise BUISSON
Laboratory
UMR 7263 IMBE - Mediterranean Institute of Biodiversity and Ecology EECAR
Project coordination
Avignon University
Partners
National Research Institute for Agriculture, Food and the Environment (INRAE) - France Centre New Aquitaine Bordeaux
Vrije Universiteit Brussel
DPT de Genetica, Ecologia e Evolução Universidade Federal de Minas Gerais
Dpt. of Plant Biology, University of Campinas
Official summary
Tropical grassland biomes (TGBs; savannahs and lawns) are ancient ecosystems with exceptional biodiversity that provide key ecosystem services.
The aim of the project is to understand the functional mechanisms that promote the establishment of herbaceous cover in TGB communities so that they are resilient to fire, sustainable, resistant to invasive alien species (IAS) and include crucial ecosystem processes. Our model is the Cerrado, the Brazilian savannah. We will be setting up a 'Biodiversity and Ecosystem Functioning' field experiment to gain a better understanding of how manipulating functional diversity and species richness can restore herbaceous communities. Until now, restoration of the Cerrado has consisted of direct sowing of fast-growing species, whereas here we will be developing methods that enable the installation of a variety of species, including conservative species, characteristic of TGBs.
1) Collection of seeds and propagules from 4 functional groups. Setting up the experiment. 2) Monitoring vegetation (surveys, functional traits, IAS, flammability) and ecosystem processes (soil erosion, productivity and carbon and water cycles). Test whether the response of functional groups to IAS is modified by extreme meteorological events, using mesocosms.
3) Implementation of controlled fires and monitoring of resilience. We expect to find a positive correlation between the complementarity of functional traits and the effectiveness of restoration. The results will contribute to deepening ecological knowledge on the assembly of TGB communities and to informing guidelines for the implementation of TGB restoration actions. Partnerships will be set up with local stakeholders and local Kalunga communities in order to optimise these guidelines.
Reference
ANR-23-CE02-0034-01
Programme
ANR - Digital sciences
Sub-programme
Artificial intelligence
Type of project
Company collaborative research project
Duration
4 years
Scientific Manager
Christèle LAGIER
Laboratory
UPR 3788 -JPEG - Legal, Political, Economic and Management Sciences Laboratory
Project coordination
University of Lorraine
Partners
CEA LIST – Paris Saclay University - Avignon University - Syllabs
Summary
The issue of diversity of information - crucial to a healthy democratic debate - has recently been raised because of the massive use of social media as a source of information. These platforms pose problems because the information presented is often unedited and selected by recommendation systems. The latter can influence users' opinions, as they tend to offer information in line with their initial opinions, trapping them in opinion bubbles. In a political context, these bubbles can lead to a polarisation of opinions, likely to generate political or public unrest. BOOM will contribute to political depolarisation by proposing new algorithms that identify and open up opinion bubbles. The project combines expertise and innovation in digital economics and media studies, political science, multimedia analysis and recommendation systems.
Reference
ANR-20-CE23-0024
Website
Programme
ANR - Cross-cutting areas
Sub-programme
Interfaces: digital sciences - humanities and social sciences
Type of project
Collaborative research project
Duration
3 years and a half
Scientific Manager
Driss MATROUF
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
Avignon University
Partners
Summary
The number of card payments has grown exponentially over the last ten years. The value
Total card transactions in 2018 amounted to almost 1,800 billion in the eurozone. The
The new European Directive PSD2, which came into force in January 2018 [1], provides for the protection of the environment.
consumers, and aims to combat fraud effectively by making it easier for them to access secure payment
electronic payments using strong authentication, also known as dual authentication.
factor. Strong authentication is based on the use of two or more elements belonging to the
knowledge" categories (what the user knows, password), "possession" (what the user knows, password), and "possession" (what the user knows, password).
(what the user has, a token) or "own" (what the user is, a biometric fingerprint, etc.).
or voice recognition). These elements must be independent in the sense that the
compromise of one does not call into question the reliability of the others, and are designed to
protect the confidentiality of authentication data.
Voice is one of the possible biometric elements for this authentication.
retained [2]. The advantage of this technique is that it is non-intrusive, can be carried out remotely and does not require the patient to be present at the time of surgery.
requires no physical contact or specialised equipment. Its reliability and simplicity
make it a perfectly acceptable identification measure in many countries.
application contexts. The aim of voice recognition is to verify a person's identity by means of their voice.
The voice is not a purely biological characteristic, but also a behavioural one:
it is not as stable and measurable as other modalities such as fingerprints or
retinal patterns. The voice is dynamic and can vary according to behaviour, age and situation.
and the state of the speaker. In addition, the characteristics of the microphones and the sound environment play an important role.
also plays an important role. This great variability is one of the greatest challenges in achieving
effective identification of the individual.
Reference
ANR-22-CE39-0009-01
Website
Programme
ANR - Specific
Sub-programme
2024 OLYMPICS FLASH
Type of project
PRCE
Duration
2 years
Scientific Manager
Johnny DOUVINET
Laboratory
UMR 7300 ESPACE - Study of Structures, Adaptation Processes and Changes in Space
Project coordination
Avignon University
Partners
ATRISC - QWANT - CHROME - GEDICOM
Summary
Cap4 Multi-Can'Alert is an experimental development project as part of the 2024 Olympic Games. Its aim is to develop an innovative multi-channel alert solution that will combine different distribution channels, adapted to the regulatory and technological contexts emerging in France, and that will integrate the needs of end-users and the reactions to be expected from the general public.
Reference
ANR-19-FLJO-0006-01
Website
Programme
ANR - Specific
Sub-programme
LABCOM CONSOLIDATION
Type of project
LABCOM CONSOLIDATION
Duration
2 years
Scientific Manager
Grégory DURAND
Laboratory
UPRI-Laboratory-Unit of Research and Innovation
ERIT-S2CB Thematic Research and Innovation Team
Project coordination
Avignon University
Partners
Summary
Since April 2015, Chem2staB, a joint research and development laboratory, has brought together teams of chemists from the University of Avignon (UMR 5247 - Equipe Chimie Bioorganique et Systèmes Amphiphiles - CBSA) and biochemists from CALIXAR. The main objective of Chem2staB is to develop new molecules capable of extracting and stabilising and/or crystallising therapeutic targets and antigens (mainly membrane proteins) without denaturing them. The search for chemical reagents for the study and validation of membrane-type pharmaceutical targets is a fairly recent field with major implications for human and animal health.
A number of studies have highlighted the essential need for the global pharmaceutical industry to innovate by stepping up its research into target/antigen validation. In particular, this means inventing new ways of preparing these targets/antigens using suitable, non-denaturing chemistry. Currently, the processes used to isolate complex targets/antigens very often use inappropriate and overly aggressive chemistry that denatures the targets/antigens from which drugs and vaccines are to be developed. This has a major impact on the performance of new drugs/vaccines.
This pooling of academic and private sector skills within Chem2staB has enabled the creation and validation of new chemical compounds with high application potential in the strategic areas of the research unit and the SME. We have marketed 11 molecules as extraction and stabilisation agents, available directly from CALIXAR and distributed worldwide by various distributors.
On the strength of its first sales in 2018, the Chem2staB laboratory is looking to consolidate its strategy of adding value to its products. The aim of the LabCom Consolidation project will be to support this consolidation phase. On the one hand, it will be necessary to promote the first molecules brought to market and enable them to be validated in applications for which few chemical solutions are currently available and for which the market potential is significant. Secondly, we will be developing new chemical agents and extractant and stabiliser formulations to counter the competition, particularly in approaches that have not yet been developed within Chem2staB.
Chem2staB's revised roadmap is now divided into 3 areas. The first involves the research and development of new detergents capable of extracting and stabilising all types of proteins (receptors, ion channels, transporters, enzymes, etc.) in their complete form (without mutation or truncation) and whatever the nature of the membranes (bacteria, viruses, yeast, mammalian cells, human cells, etc.).
A second area focuses on the biochemical validation of the various molecules developed on targets/antigens of scientific and industrial interest produced by CALIXAR on its own behalf or on behalf of third parties, particularly in industry but also in the academic life sciences.
The third area concerns the scientific and economic exploitation of the reagents validated in area 2. Some reagents feed directly into the pipeline of compounds used by CALIXAR to isolate/formulate targets and antigens, while others are marketed through a network of distributors worldwide.
Reference
ANR-18-LCCO-0003-01
Website
Programme
Horizon Europe
Sub-programme
HORIZON-CL4-2023-DATA-01
Type of project
Duration
Scientific Manager
Francesco De Pellegrini
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
Centre for Research and Technology-Hellas (GR)
Partners
FIWARE FOUNDATION EV (DE) - SBA Research Gemeinnutzige Gmbh (AT) - National Center for Scientific Research Demokrit (EL) - Fraunhofer Gesellschaft Zur Forderung Der Ang (DE) - Teknologian Tutkimuskeskus vtt Oy (FI) - University of Lancaster (UK) - Elliniko Mesogeiako Panepistimio (EL) - Netcompagny-Intrasoft SA (LU) - Engineering Ingegneria Informatica SPA (IT) - Siemens SRL (RO) - AVL List Gmbh (AT) - Axon Logic Idiotiki Kefalaiouxiki Etaireia (EL) - Beyond Semiconductor, Raziskave In Razvoj DOO (SI) - K3Y (BG) - Ubitech Limited (CY)
Summary
CoGNETs aims to revolutionise intelligent infrastructure management by introducing a middleware framework
scalable, interoperable distributed middleware for autonomous IoT-to-Cloud computing, sustainable during and after the project via the FIWARE Foundation
(Technical Director) and supported by a solid industrial and academic ecosystem
between the EU and Japan on data sovereignty for the automotive and manufacturing supply chain.
The idea is to leverage computational intelligence to the point where devices can realise their heterogeneity themselves and decide how to form a dynamic IoT-to-Cloud swarm to respond to common AI tasks and important elements of cognitive computing in an automated, secure and energy-efficient way.
At the heart of our middleware, we will integrate a new federated and decentralised multi-context broker architecture, enriched with intelligent game agents, federated collaborative learning and an information management system (including RISC-V hardware security and AI acceleration) to perform dynamic and cognitive computing tasks.
CoGNETs will also achieve energy/CO2 neutrality and security awareness "by design" by leveraging all of its middleware processes for unified IT, energy and security optimization, introducing a new logic for systematically improving IT jointly and equally with the energy and security performance needed to support new services and business models.
This new logic makes it possible to systematically improve IT in terms that are joint and equal to the energy and security performance required to support new services and business models.
Validations will be carried out on 3x vertical demos and 1x cross-vertical demos to support the emerging sectors of manufacturing, mobility and health and to examine how CoGNETs can safeguard economic, commercial and societal resilience, and how it is possible to extend Europe's digital sovereignty, strategic autonomy and industrial security in the next wave of technological evolution.
Reference
101135930
Programme
ANR-Life Sciences
Sub-programme
Biochemistry and life chemistry
Type of project
Collaborative research project
Duration
4 years
Scientific Manager
Grégory DURAND
Laboratory
UPRI-Laboratory-Unit of Research and Innovation
ERIT-S2CB Thematic Research and Innovation Team
Project coordination
Avignon University
Partners
Montpellier University, I2BC, CALIXAR
Summary
The originality of Cytergents' proposal lies in the development of detergents inspired by cyclic and lipid metabolites.
for extracting and stabilising PMs, thereby avoiding their exposure to the more conventional, more aggressive detergents. From a chemical point of view, the originality of the synthesis lies in the use of C-C bond formation for the preparation of detergents and the use of a lipid metabolite from a lipid-metabolite platform. We envisage that the detergents
identified cyclic detergents can be used as mild detergents to extract PM directly from the host membrane, offering a decisive advantage over most other systems
which continue to depend on detergents for the solubilisation and purification of PM. for solubilisation and purification.
Programme
ANR-Specific
Sub-programme
Water4All
Type of project
PRCI
Duration
3 years
Scientific Manager
Konstantinos CHALIAKIS
Laboratory
EMMAH
Project coordination
Vanvitelli University
Partners
University of the Western Cape (ZA) - Instituto de Geociência, Universidade de São Paulo (BR) - Politechnika Gdanska (PL)
Summary
The overall aim of DATASET is to develop an effective tool to help protect and regulate the use and management of water resources in coastal aquifers, taking into account all the dynamic conditions caused by the predicted increase in the frequency of HWT.
In this way, we will promote the protection of groundwater, a resource of unfathomable value to the region's many inhabitants.
We are therefore going to promote the protection of groundwater, which is an unfathomably valuable resource for the many human activities and fragile environments found in coastal plains.
DATASET aims to achieve the following objectives:
OBJ1: To develop and apply a method for creating vulnerability maps through a holistic assessment of coastal aquifers, coupling the assessment of AGR and SP in a single methodology in order to improve water governance for early warning, prevention and mitigation of groundwater quality degradation under current and future conditions, taking into account the impacts of climate change and water pollution.
OBJ2: To promote a paradigm shift in water management, thanks to access to data and models via the Internet,
the dissemination of management tools for planning and community involvement and their transfer to stakeholders.
OBJ3: To promote social awareness of the importance of groundwater and its sensitivity to climate change and water pollution. From stakeholders and decision-makers to end-users.
Reference
ANR-23-W4AP-0001-04
Programme
ANR-Cross-cutting areas
Sub-programme
Digital revolution
Type of project
Collaborative research project
Duration
5 years and 3 months
Scientific Manager
Pierre-Henri MORAND and Vincent LABATUT
Laboratory
LIA/LBNC
Project coordination
Avignon University
Partners
Centre de recherches administratives - DATACTIVIST
Summary
The massive opening up of public data is of considerable economic and social importance. This is particularly true of public procurement data, which is now available, and on which hopes are founded of discovering and combating the phenomena of fraud and corruption that are unfortunately present on a massive scale; and this by making it possible to highlight critical information and develop tools to improve the effectiveness of the law. Combining information technology, economics and law, DeCoMap aims to collect, process and analyse data relating to French public procurement contracts, in order to develop tools for the automatic detection of corruption and fraud risks and to propose a normative analysis grid highlighting the main risk factors that legislators should identify and to which the supervisory authorities should turn their attention. Supported by Transparency International France and the Open Contracting Partnership, DeCoMaP brings together 10 lecturers and researchers from 7 universities specialising in public procurement law and data law, economic and econometric analysis of public contracts, economic analysis of law, graph optimisation, and analysis and extraction of complex networks. 4 members of Datactivist, a co-operative specialising in Open Data and heavily involved in the opening up of public procurement data, complete the consortium.
The project aims to establish and analyse a comprehensive database of identified corrupt practices, drawing on a variety of primary and secondary legal sources.
Collecting the data required for the project is a major undertaking, and will make it possible to create the first database on fraudulent practices in French public procurement. The project will make it possible to establish and analyse this ground truth in economic and legal terms.
Reference
ANR-19-CE38-0004-02
Website
Programme
ANR-Générique
Type of project
PRC
Duration
4 years and 6 months
Scientific Manager
Yannick Estève
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
INRIA Grand Est
Partners
University of Le Mans Computer Science Laboratory - Magnet
Summary
Speech recognition is now used in many applications, including virtual assistants that collect, process and store personal speech data on centralised servers, raising serious privacy concerns. Approaches based on embedded speech recognition have recently been proposed to deal with these privacy aspects, but only during the speech recognition phase. In this case, as all processing is carried out on the user's terminal, speech data remains private. However, there is still a need to further improve speech recognition technology, as its performance remains limited under unfavourable conditions (e.g. noisy environments, reverberated speech, strong accents, etc). This can only be achieved using large speech corpora representative of real and varied conditions of use. To achieve this, it is necessary to share speech data while keeping the identity of the speaker private. The improvements are then beneficial for all users. It is also clear that the user must have control over his data, so as not to transmit data whose linguistic content is critical.
In this context, DEEP-PRIVACY proposes a new paradigm based on a distributed, personalised and privacy-preserving approach to speech processing, with a focus on learning algorithms for speech recognition. To this end, we propose a hybrid approach: each user's terminal does not share its raw speech data and performs private computations locally, while some inter-user computations are performed on a server (or peer-to-peer network). To meet privacy requirements, the information communicated to the server must not expose sensitive information. The project addresses the above challenges from a theoretical, methodological and empirical point of view through two major scientific objectives.
The first objective concerns the learning of privacy-preserving representations of the speech signal, i.e. those that disentangle characteristics likely to expose private information (to be kept on the terminal) from generic information useful for the task in question (which satisfies aspects of privacy, and can be shared). For speech recognition, this corresponds respectively to speaker information (to be protected) and linguistic information (to be shared) carried by speech. To achieve this goal, we will explore several directions, all based on deep learning approaches; and, in addition to classical speech and speaker recognition measures, we will also use formal notions of privacy to evaluate their performance.
The second objective concerns distributed algorithms and personalisation, through the design of efficient distributed algorithms operating in an environment where sensitive user data is kept on the terminal, with global components running on servers and personalised components running on personal terminals. The data transferred to the servers should contain useful information for learning and updating the global components (acoustic models), while preserving privacy. We will study the type of data to be exchanged (e.g., gradients, partial models, etc.) and the speaker information remaining in this data. In addition, custom components can be used to introduce speaker-specific transformations and adapt certain model parameters to the speaker. Finally, we will consider a peer-to-peer context, as an alternative to servers, for data sharing and model learning.
Reference
ANR-18-CE23-0018-04
Website
Programme
ANR - Digital sciences
Sub-programme
Artificial intelligence and data science
Type of project
Collaborative research project
Duration
4 years and 1 month
Scientific Manager
Rachid ELAZOUZI
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
Avignon University
Partners
LAAS-CNRS, Toulouse - University of Toulouse – New York University (United States) - University of Oulu (Finland)
Summary
Energy efficiency is gaining ground in the AI community as researchers are made aware of the importance of
carbon as well as the dollar costs of the current race to improve the accuracy of machine learning
applications. Recent work has highlighted these hidden costs of existing state-of-the-art models
for natural language processing (NLP) [St19] and image recognition [Sc20]. In [Sc20], it was shown that
the training of a single NLP model emitted 300,000 kg of CO2, the equivalent of 125 return flights between New York and New York.
York and Beijing. One of the most elaborate deep learning models, known as GPT-3, requires an amount of
equivalent to the annual energy consumption of 126 Danish households. If the growth of machine learning
applications continue in the current direction, the amount of energy required for machine learning will be
quickly become prohibitive from a climatic, technical and economic point of view. With access to large
large-scale data resources, the number of calculations used to train learning models has increased by 300,000 in 6
years [Sc20]. What's more, massive computing not only has a very high carbon footprint, but it also
negative effects on the inclusion of research and the deployment of real AI-based applications, particularly in
the current context of rising energy prices and new record highs.
Reference
ANR-22-CE23-0024-01
Website
Programme
Sub-programme
FLASH OURAGAN
Type of project
Collaborative research project
Duration
4 years and 4 months
Scientific Manager
Damien SERRE
Laboratory
UMR 7300 ESPACE - Study of Structures, Adaptation Processes and Changes in Space
Project coordination
Partners
Urban Planning Laboratory - Espaces et sociétés - France IFSTTAR - Association, Robin des Bois - Study of structures, adaptation processes and changes in space
Summary
Initial reports from the areas affected by Hurricanes Irma and Maria show that the quantities of waste have largely disrupted waste management services (UN Environment / OCHA Joint Unit, 2017). In October 2017, the Grandes Cayes Ecosite in Saint-Martin received the equivalent of two and a half years' worth of normal-period waste collection. These figures, common in disasters of this type, have already been observed in the past. Feedback also shows that waste management is an important issue in the post-crisis period. According to the US Federal Emergency Management Agency (FEMA), the cost of post-disaster waste management is more than 1/3 of the cost of restarting the area. It is also a question of image for the areas affected, particularly for the tourist industry. The health and environmental implications of poor waste management can have an impact on the health of local residents and pollute soil and water. Finally, the large amount of waste also hinders the progress of emergency services and the access of technical services to drinking water collection and production facilities, electrical transformers, etc. International experience shows that improving the management of post-disaster waste can also improve the management of everyday waste in normal situations. The DéPOs project focuses specifically on the problem of post-hurricane waste management. It provides an opportunity to work with experienced partners and to bring together a wide range of skills, approaching the problem from complementary angles: predictive quantification of deposits and characterisation (axis 1), integrated waste management (axis 3) and spatial modelling (axis 2). DéPOs is focusing on applied research to build the methodological and specific knowledge required for improved post-disaster management. DéPOs also proposes research that is involved in the areas concerned. On the basis of the knowledge acquired in Areas 1 and 2, the fieldwork in the French West Indies gives rise in Area 3 to management proposals built in close collaboration with local stakeholders. The regional departments (Martinique and Guadeloupe) responsible for planning and managing post-hurricane waste in the French West Indies have confirmed their willingness to collaborate.
In the scientific field, DéPOs contributes to the field of research on spatial risks and post-crisis resilience. Beyond the hazard/vulnerability binomial, DéPOs provides knowledge on the implementation of resilience through planning, the adaptation of organisations and the functioning of socio-technical systems subject to exceptional constraints.
In normal times, waste management is an important service for the functioning of local authorities, but it becomes essential after a hurricane.
Reference
ANR-18-OURA-0003-03
Website
Programme
AAP - Generic, digital sciences
Sub-programme
Artificial intelligence
Type of project
JCJC
Duration
3 years and 6 months
Scientific Manager
Richard DUFOUR
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
Avignon University
Partners
Summary
A major problem with evaluation metrics in language processing is that they are designed to measure an approach globally against a benchmark, and thus compare systems against each other. Although these automatic systems are intended for end-users, these users are little studied: the impact of these automatic errors on humans, and the way in which they are perceived at a cognitive level, has never been studied, and is therefore absent from the evaluation process. The DIETS project proposes to focus on the problem of diagnosis/evaluation of end-to-end automatic speech recognition (ASR) systems by integrating the human reception of transcription errors from a cognitive point of view. The challenge is twofold: 1) to analyse PAR errors in detail in relation to human reception and 2) to understand and detect how these errors manifest themselves in an end-to-end PAR framework, the work of which is inspired by the human brain.
Reference
ANR-20-CE23-0005
Website
Programme
ANR- Human and Social Sciences
Sub-programme
Societies and territories in transition
Type of project
International collaborative research project
Duration
3 years
Scientific Manager
Laure CASANOVA
Laboratory
UMR 7300 ESPACE - Study of Structures, Adaptation Processes and Changes in Space
Project coordination
Avignon University
Partners
Université Polytechnique Hauts-de-France - LARSH
Summary
There is renewed interest in the distribution of property wealth and its socio-economic implications. The current context is marked by increasing inequality and a more flexible tax and regulatory regime. The project proposes to broaden the field of study of political economy by assessing the socio-legal embedding of land markets in France and Luxembourg, using two case studies: the metropolises of Aix-Marseille and Luxembourg. The DISTRILAND project analyses the spatial impact, in terms of housing production and planning, of the interrelationships between inherited land structures, regulatory systems and institutional structures.
Reference
ANR-22-CE55-0009-01
Website
Programme
ANR- Mathematics and its interactions
Sub-programme
Mathematics
Type of project
Collaborative research project
Duration
5 years
Scientific Manager
Erwann DELAY
Laboratory
UPR 2151 LMA - Mathematics Laboratory
Project coordination
Sorbonne University
Partners
LMA (AU), Institut Denis Poisson (Université François Rabelais Tours)
Summary
Parametrisation of the initial data set of Einstein's equations. The project focuses on the global geometry of Riemannian varieties satisfying the Einstein constraints appearing in general relativity. In other words, the project is interested in the geometric and analytic properties of data sets, consisting of a space-like hypersurface (representing a slice of "present" time) in a space-time satisfying the Einstein field equations (possibly coupled to matter fields). Among our main objectives, we will seek a parametrisation of 'all' these hypersurfaces and describe their global geometric and asymptotic properties, for example their behaviour at infinity as well as in space or in the vicinity of gravitational singularities.
Reference
ANR-23-CE40-0010-02
Programme
ANR - Generic
Type of project
PRC
Duration
4 years and 9 months
Scientific Manager
Frédéric MONIER
Laboratory
Project coordination
Partners
Techniques, Territories and Societies Laboratory - TRIANGLE - Centre Universitaire de Recherches sur l'Action Publique et le Politique - Centre de sociologie des organisations - Centre de recherches juridiques - Centre d'études et de recherches administratives, politiques et sociales
Summary
In France, the financial remuneration of politicians, whether elected at local or national level, represents an expense of over a billion euros a year. These costs associated with political work are regularly the subject of fierce criticism in a more or less accusatory vein. According to this view, paid elected representatives are expensive and put their personal financial interests first. The ELUAR project seeks to break away from these common, all-encompassing representations by examining in detail the role played by financial rewards in the process of professionalising elected representatives. From a scientific point of view, it seeks to fill a gap in the French literature on political work. Although there has been a proliferation of publications on the subject since the 1990s, analysis of the material conditions under which mandates are exercised remains a blind spot in French research. Taking an interdisciplinary approach (sociology, political science, history, law), this collective research project aims to reintroduce the financial dimension into the analysis of the careers and commitments of elected politicians. The central hypothesis of the project is to highlight the heterogeneity and inequality of the remuneration of elected representatives and the forms of political professionalisation. In practical terms, the project is structured around two strands. The first involves studying the production of reforms and the legal framework in order to reveal the political construction of an economic hierarchy between mandates. Who are the actors involved in producing the reforms? What justifications have been used since the 1950s? What role does the principle of multiple mandates play in these games in France? What remuneration and material rewards are available to elected representatives? Are the same hierarchies found in other countries? The second section looks at how the rules governing the remuneration of elected representatives are used and appropriated. Focusing on remuneration and, more broadly, on the material conditions under which elected office is held will provide an insight into the variety of contemporary forms of political professionalisation, the subjective relationship between elected representatives and money, and the political uses to which money is put. How do elected representatives manage to abandon their initial profession in favour of a political mandate? What economic security strategies do they deploy? How are financial allowances and bonuses actually allocated to elected representatives? Can money be used to secure the loyalty of political teams? Is it used as a political weapon to disqualify opponents? These are just some of the questions that this second section will address. In the final analysis, the ELUAR project seeks to make a double break. A break with the ordinary discourse that tends to homogenise elected representatives and assess their remuneration in terms of suspicion, and a break with the scholarly point of view that emerges in research on political work, according to which compensation almost automatically makes a professional.
Reference
ANR-16-CE26-0013-02
Website
Programme
Horizon 2020
Sub-programme
H2020-MSCA-RISE
Duration
4 years
Scientific Manager
Jean-François Bonastre
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
Partners
Laboratoire national de metrologie et d'essais (FR) - Vysoke Uceni Technicke V Brne (CZ) - Université Grenoble Alpes (FR) - Avignon Université (FR) - University of Sheffield (UK) - Allo-Media (FR) - ELYADATA (TN) - Omilia Ltd (EL) - Phoenexia Sro (CZ) - University of Yaounde (CM) - Consejo nacional de investigaciones cientificas y tecnicas (AR) - Universidad de Chile (CL) - Centro de aplicaciones de tecnologias de Avanzada (CU) - Universiti Malaysia Sarawak (MY) - Universiti sains Malaysia (MY) - John Hopkins University (US) - MILA (CA)
Summary
The ESPERANTO project aims to take speech processing technologies to the next level, enabling them to be disseminated among European SMEs and maximising and securing their use in civil society for forensic medicine, healthcare and education.
The ESPERANTO consortium predicts that the next generation of artificial intelligence algorithms for speech processing should :
1. be more accessible: via a greater number of spoken languages, and for applications where resources are severely limited (health, education, robotics).
resources are severely limited (health, education, robotics, etc.);
2. bring a human into the loop to ensure greater ease of use, deployment and maintenance;
3. be explainable to enable sensitive forensic or health applications and contribute to the preservation of personal data by detecting and characterising existing biases due to the data-driven nature of current speech technologies.
ESPERANTO intends to guide the scientific community by publishing evaluation measures, protocols and standards that will make it possible to
to stimulate the development and evaluation of this new generation of algorithms. To achieve this ambitious goal, the ESPERANTO project is bringing together a large, cross-sector community of experts in speech-related applications, such as transcription, separation, enhancement, translation, understanding and speaker, in order to transfer knowledge, organise, produce and standardise resources.
The aim is to catalyse and cross-fertilise efforts in this area.
catalyse and cross-fertilise efforts in this area.
The main objectives of the ESPERANTO project are as follows:
- support the development of open-source tools that encourage rapid development, exchange and reproducibility;
- Produce tutorials and competitive baselines on various speech processing topics to encourage the creation of new students, researchers and engineers in speech AI;
- facilitate the collection and sharing of linguistic and speech resources through standards ;
- organise workshops to advance speech technologies and promote knowledge transfer.
Reference
101007666
Programme
ANR - Digital sciences
Sub-programme
Artificial intelligence and data science
Type of project
Collaborative research project
Duration
4 years
Scientific Manager
Yannick ESTEVE
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
Avignon University
Partners
Paris Dauphine University – Grenoble Computer Science Laboratory
Summary
Self-Supervised Learning (SSL) has recently emerged as an incredibly promising artificial intelligence (AI) method. Thanks to this method, the colossal masses of unannotated data that are accessible can be used by AI systems to surpass previously known performances. In particular, the field of automatic speech processing (ASP) is being rapidly transformed by the arrival of SSL, thanks in particular to massive industrial investment and the explosion of data, both made available by a handful of companies. The gains in performance are impressive, but the complexity of SSL models means that researchers and manufacturers in the sector have to equip themselves with unprecedented computing capacity, drastically reducing both access to fundamental research in this area and its deployment in everyday products. For example, much of the work using an SSL model for TAP is based on a system maintained and made available by a single company (wav2vec 2.0). The entire life cycle of the technology, from its theoretical foundations to its practical deployment, including analysis of the societal aspects, therefore depends solely on institutions with the physical and financial resources to support the intensive development of this technique. The E-SSL project aims to give the TAP scientific community and industry the necessary control over self-supervised learning to ensure its evolution and equal deployment by facilitating both academic research and its transfer to industry. In practice, E-SSL holistically integrates three key issues of self-supervised learning for TAP, including its effective computational efficiency, its societal impact and the feasibility of extending it to tomorrow's products.The BRUEL project concerns the evaluation/certification of voice identification systems against adversarial attacks.
Reference
ANR-22-CE23-0013-01
Website
Programme
Digital sciences
Sub-programme
Artificial intelligence and data science
Duration
4 years
Scientific manager
Jean-François BONASTRE
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
Orange
Partners
INSTITUT DE RECHERCHE ET COORDINATION ACOUSTIQUE MUSICALE
UNIVERSITY OF RENNES
Centre National de la Recherche Scientifique (CNRS) - Ile-de-France Regional Office -Villejuif
Official summary
Describing a voice in a few words is an arbitrary task. We can talk about a 'deep', 'blown' or 'husky' voice, but characterising a voice would require a restricted set of rigorously defined attributes constituting an ontology. But no such description grid exists. Machine learning applied to speech suffers from the same weakness: in most automatic processing tasks, the speaker is modelled by abstract global representations with characteristics that are not, or only slightly, explicit. For example, automatic speaker identification is generally approached using the x-vector paradigm, which consists of describing a speaker's voice using an embedding specially designed for this task. Despite their good accuracy, x-vectors are generally unsuitable for detecting similarities between different voices with common characteristics. The same observations apply to speech generation: speech synthesis is generally controlled by injecting the speaker's style or identity via unstructured representations. These representations make it possible to bypass the task of defining and learning ontologies, but they only make it possible to imitate a subset of a voice's characteristics (genre, fundamental frequency, rhythm, intensity) without making its attributes explicit. They are also limited by their inability to generate new, original voices. The aim of this project is to decipher the codes of human voices by learning explicit and structured representations of voice attributes. Achieving this objective will have a strong scientific and technological impact, in at least two areas of application: firstly, in speech analysis, it will enable us to understand the complex tangle of characteristics of a human voice; secondly, for voice generation, it will feed a wide range of applications for creating a voice with the desired attributes, enabling the design of what is known as a vocal personality. The set of attributes will be defined by human expertise or discovered from the data using lightly supervised or unsupervised neural networks. It will include a detailed and explicit description of timbre, voice quality, phonation, speaker biases such as specific pronunciations or speech impairments (e.g. lisping), regional or non-native accents, and paralinguistic elements such as emotions or style. Ideally, each attribute could be controlled in synthesis and conversion by a degree of intensity, allowing it to be amplified or erased from the speech, as part of a structured integration. These new attributes could be defined by experts or by neural network algorithms such as automatic voice unravelling or self-supervised representations that would automatically discover salient attributes in multi-speaker datasets. The main industrial results expected concern different use cases for voice transformation. The first is voice anonymisation: to enable RGPD-compliant voice recordings, voice conversion systems could be configured to remove attributes strongly associated with a speaker's identity, while other attributes would remain unchanged to preserve the intelligibility, naturalness and expressiveness of the manipulated voice;
the second is voice creation: new voices could be sculpted from a set of desired attributes, to feed the creative industry.
Reference
ANR-23-CE23-0018-04
Website
Programme
ANR-Mounting of European or International Scientific Networks
Sub-programme
Type of project
Setting up European or international scientific networks
Duration
2 years
Scientific Manager
Francesco DE PELLEGRINI
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
Avignon University
Partners
INRIA (France)
TU Delft (Netherlands)
Technion (Israel)
IMDEA (Spain)
NOKIA (France)
Summary
FINALITY in a Nutshell. FINALITY is a MSCA DN forming a novel AI curriculum for engineering researchers
exploring safe techniques for socio-technical systems where human decisions for resource allocation are
supported by AI. The ESRs engaged in the FINALITY DN will develop new methodological tools focusing on the
following expertise areas: constrained and delayed MDP theory and their application to safe Reinforcement
Learning, Online Convex Optimization and Federated Learning under constraints
Reference
ANR-23-MRS1-0007-01
Website
Programme
ANR - Generic
Type of project
PRCI
Duration
4 years and 1 month
Scientific Manager
Grégory DURAND
Laboratory
UPRI-Laboratory-Unit of Research and Innovation
ERIT-S2CB Thematic Research and Innovation Team
Project coordination
Avignon University
Partners
Institute of Structural Biology - Martin Luther University Halle-Wittenberg - University Kaiserslautern
Summary
Membrane proteins play an essential role in cellular communication and transport processes and therefore represent the majority of current therapeutic targets. Unfortunately, progress in our knowledge and understanding of their structures, dynamics and functions is currently insufficient, mainly because of experimental difficulties due to their highly hydrophobic nature.
In order to manipulate them in vitro and keep them both soluble and active, these proteins require an environment that mimics the membrane. Traditionally, this has been achieved using surface-active compounds, known as detergents, which are capable of extracting them from the membrane and then keeping them soluble. However, many proteins become denatured once extracted by these same detergents, and lose their native structure and function.
These observations have motivated research aimed at proposing gentler alternatives to conventional detergents, among which fluorinated surfactants seem particularly promising. Due to the low affinity of perfluorinated segments to hydrogenated segments and the greater volume occupied by fluorinated chains compared to hydrogenated chains, fluorinated surfactants are considered to be less denaturing than conventional detergents. Many examples report increased stability of membrane proteins solubilised in fluorinated surfactants. This is due to the fact that they compete less with protein-protein and protein-lipid/hydrophobic cofactor interactions, which stabilise the protein structure. However, the main disadvantage of using fluorosurfactants lies in the need to extract the protein from the membrane using conventional detergents, then transfer it to fluorosurfactants at a later stage in the purification process. Furthermore, by this stage, the most sensitive proteins have already suffered irreversible damage.
We have very recently demonstrated that, contrary to common belief, the fluorinated nature of the hydrophobic parts of fluorosurfactants does not compromise their ability to dissolve artificial membranes. Fluorinated surfactants could therefore have both extracting and stabilising properties in relation to membrane proteins.
In this project, our aim is to continue our efforts to develop and test fluorinated detergents (i) that can be synthesised in sufficient quantity and purity to allow rapid use in the scientific community; (ii) capable of displaying self-assembly properties compatible with structural analyses of membrane proteins using X-ray crystallography in particular; (iii) also capable of partitioning and solubilising lipid bilayers rapidly and in a thermodynamically controlled manner; (iv) extracting membrane proteins directly from native or synthetic membranes without the need to add detergents; and (v) providing a stabilising environment for the proteins thus extracted so that they retain their native structure and function over time ranges compatible with their functional and structural study.
Such fluorinated detergents would open up new possibilities for the in vitro study of membrane proteins of physio- and pharmacological interest that are too unstable in detergent to be studied. This highly interdisciplinary project will be carried out by a Franco-German consortium that is particularly well qualified to successfully handle the synthesis, physico-chemical, biophysical and biochemical studies, as well as the structural biology of fluorinated detergents and membrane proteins solubilised in this new medium.
Reference
ANR-16-CE92-0001-01
Website
Programme
ANR - Generic
Type of project
PRC
Duration
3 years and a half
Scientific Manager
Georges LINARES and Damien MALINAS
Laboratory(ies)
UPR 4128 LIA - Avignon Computer Laboratory
UMR 8562 CNE - Dynamics of Social Worlds ECC
Project coordination
Avignon University
Partners
Syllabs - Research and polling firm - EURECOM
Summary
Festivals are becoming increasingly popular, and some of them, such as Cannes and Avignon, are generating very significant activity on the Web. The project aims to analyse this activity in order to gain a better understanding of festival practices and develop methods for accessing and displaying the content generated in and around festivals. These two objectives are complementary, and the underlying scientific problems involve both information sciences and technologies and the humanities and social sciences, which are combined in the project.
The project therefore focuses on two areas: the study of uses via data collected on the web, and the re-editorialisation of content captured or produced by web users. These two areas are based on the collection of festival-related data from a variety of sources (Twitter, blogs, forums, etc.). Structuring these collections should make it possible to carry out surveys, extract knowledge about audiences and cultural practices, and show particular points of view targeting a particular category of spectator:
- structuring: this will involve extracting descriptors, identifying similarities and producing concise representations of content. We will propose robust document representation paradigms, based in particular on partial regression methods, which combine a priori knowledge and unsupervised analysis of large quantities of data.
- digital terrain: the aim of this area is to rethink what could be the tools for surveying and monitoring audiences in their choices and their spectator interactions (choice of a show, advice, evaluation). The invention of a digital field raises the question of the protocol for observing audience behaviour, the quantity of information, the strategies for collecting and selecting it, and the ability to process what is to become data. The automatic structuring of the data collected will therefore lead us to address the question of the existence of digital terrains and to propose a technological and methodological framework for their exploration.
-re-editorialisation: the aim is to use the video data collected to create summary views of the collections. In concrete terms, this will involve automatically producing films to meet a specific user request (for example, a short presentation of the shows taking place during a day at the Transmusicales). We will be evaluating approaches based on extracting and then reassembling video segments, while addressing fundamental issues relating to functions of interest in a cultural context and compositional logics, open problems that will be tackled jointly by computer scientists and sociologists.
The project's potential for breakthrough is therefore linked to the answers it will provide to the question of digital research fields and the way in which we envisage integrating knowledge of audiences and the reception of works into video content re-editorialization systems. The experiments will be carried out at 4 major festivals: the Avignon theatre festival, the Lumière festival, the Transmusicales and the Cannes festival. The project was presented to the organisers of the first three festivals and to UniversCiné, which covers the Cannes festival to a large extent. The results of the project will be exploited through the development of a 'Festival Observatory', an open application aimed at the public, organisers and analysts alike. This application should offer a synchronous view of the indicators extracted automatically, enable navigation through the data collected and structured, and show video previews composed automatically according to a user request.
Reference
ANR-14-CE24-0022-01
Website
Programme
ANR - Generic
Type of project
PRC
Duration
3 years and 11 months
Scientific Manager
Grégory DURAND
Laboratory
UPRI-Laboratory-Unit of Research and Innovation
ERIT-S2CB Thematic Research and Innovation Team
Project coordination
CNRS Languedoc-Roussillon
Partners
Bioorganic Chemistry and Amphiphilic Systems Team
Summary
G protein-coupled receptors (GPCRs) are one of the main families of membrane receptors and are involved in many fundamental cellular processes. They are therefore one of the main targets for therapeutic molecules, with potential applications in a wide range of clinical areas, including neurological and metabolic disorders, inflammation, cancer and viral infections. Most research programmes for drug candidates targeting a GPCR rely on the use of screening systems based on cells overexpressing the receptors, followed by detection of ligand binding or production of second messengers. Although these programmes have led to the identification of molecules currently used in therapeutics, the screening techniques employed have major drawbacks that limit the identification of original compounds. In particular, the new concepts emerging from GPCR pharmacology, such as functional selectivity, receptor dimerisation and allosteric response modulation, raise the problem of choosing an appropriate detection method. Similarly, the screening of low-affinity fragments is impossible in cellular systems, even though it is a central process in medicinal chemistry. Access to assays that can visualise multiple signalling pathways, take account of receptor dimerisation and ligand functional selectivity, and enable fragment screening is therefore absolutely essential for the identification and optimisation of novel compounds targeting GPCRs. In this context, we are proposing a multidisciplinary programme aimed at developing original biosensors for cell-free screening tests based on the immobilisation, on solid supports, of receptors purified in the form of monomers and heterodimers. Immobilisation will be achieved using functionalised non-ionic amphiphilic polymers, enabling the native protein to be bound. These biosensors have most of the characteristics required for a high-throughput test, namely the ability to screen a large number of compounds, the preservation of the pharmacological properties of the receptors after immobilisation, low implementation costs, the ability to miniaturise and automate the test, the absence of the use of radiolabelled compounds, and sensitivity and robustness. They will be validated by screening banks of compounds using two complementary techniques. The first, Surface Plasmon Resonance, will provide information on the binding properties of ligands on the immobilised receptor (affinity, kinetic constants). The second, based on the use of fluorescence, will provide information on the state of activation of the receptor under the effect of ligand binding. The proof of concept will be demonstrated with the ghrelin GHSR receptor. In addition to the fact that this receptor is a GPCR model allowing subsequent extension to other receptors in the same family, it is also a major therapeutic target with potential applications in the treatment of obesity, diabetes and addiction to drugs of abuse or alcohol. This work should thus, on the one hand, enable the implementation of original biosensors for screening compounds targeting GPCRs and, on the other hand, the identification of original ligands of the GHSR receptor for subsequent therapeutic use.
Reference
ANR-17-CE18-0022-03
Website
Programme
ANR - Specific
Sub-programme
FLASH DATA
Type of project
PRC
Duration
2 years
Scientific Manager
Jean-François BONASTRE
Laboratory
UPR 4128 LIA - Avignon Computer Laboratory
Project coordination
Partners
Inria Nancy Grand-Est Research Centre
Summary
With the increasing use of voice interfaces and smartphone applications, a growing amount of voice data is being captured, stored and used by service providers. In the majority of cases, this use of voice data shows no malicious intent.
However, by its very nature, voice data contains personal, sensitive information that should not be passed on to others.
They provide information about your state of health, your socio-economic status, your geographical and ethnic origins, your personality or your emotional feelings. Voice recordings are also a source of information about your family circle, close friends and professional relationships.
Protecting this information from malicious or ethically reprehensible use is a necessity to extinguish the risk of breaches of our privacy.
There are two strategies for achieving this: protecting/encrypting data or anonymising it. While the choice of the best solution naturally depends on the intended application, these two strategies are complementary, on the one hand, and anonymisation is more flexible at a lower additional cost, on the other.
Anonymisation techniques can be used to remove the personal elements to be protected from the voice signal, while maintaining the intelligibility and quality of the message. Once anonymised, voice recordings can be processed, stored and (re)used without the risk of linking the data elements to the speakers concerned. Unfortunately, the range of anonymisation solutions on offer and the progress they have made remain limited by a lack of tools and open data sets accessible to all. These elements are essential for evaluating the performance of solutions and comparing them. As with any pattern recognition problem, this lack of resources is a major obstacle.
Harpocrates will form a working group and a resource development community which, as well as offering the first open resources in the field, will also organise the first international challenge in speech data anonymisation. Experience gained in other fields (speech recognition, speaker recognition, language recognition, etc.) shows that this kind of effort, pursued year after year, leads to significant progress and rapid transfer to industry. This last point is crucial, as demand for personal data protection is growing strongly and urgently (published data can no longer be protected). Anonymisation solutions will in fact be a necessary component in meeting the expectations of legislation on the protection of sensitive data, for 'privacy by design' development approaches, and so on.
Reference
ANR-19-DATA-0008-02
Website
Programme
ANR - Specific
Sub-programme
SHS FRAL
Type of project
PRCI
Duration
4 years
Scientific Manager
Frédéric MONIER
Laboratory
UMR 8562 CNE Dynamics of Social Worlds HEMOC
Project coordination
Avignon University
Partners
Technische universität Darmstadt Institut für Geschichte - Sorbonne identités
Summary
Transparency is now seen as a political value, synonymous with democracy, participation and accountability. This overlooks two points: demands for transparency are neither temporary nor genuinely new, but are linked to specific contexts and have their own history. What's more, they have ambivalent effects. As there are as yet virtually no historical studies of transparency, we have focused on political history.
Apparently, transparency is a property of a political system. In reality, it is expressed in political demands for transparency from below, particularly in terms of access to information and issues that politics makes "readable". We intend to focus on the history of demands for transparency in politics in Germany and France. Our lines of enquiry concern, firstly, the history of parliamentary committees of enquiry between 1890 and around 1970, and secondly, two scandals that had a major impact on party funding (the Flick scandal and the Urba affair). This raises a number of questions: when did transparency become an effective requirement? What measures were proposed? Who were the main players? In what configurations? Did the demand for transparency lead to greater or lesser awareness of political processes? Has it strengthened or reduced political trust? Among the theories of transparency, the analytical categories developed by David Heald stand out [Heald, 2006].
One of our working hypotheses is that, in the first half of the twentieth century, demands for transparency tended to focus on individual members of the political staff, on a case-by-case basis. From the 1970s-1980s, the transparency injunction took on a new dimension, and this demand was directed at the system itself, which was required to be transparent. In support of this interpretative hypothesis, one argument can be put forward: the growing demand for transparency has not strengthened confidence in democracy. Moreover, the general configuration and the players have changed. From the 1980s onwards, new players from civil society began to appear alongside the press, calling for instruments that would produce transparency.
The project proposed here is part of a cooperation between French and German researchers who have been working together since 2011 on historical studies of corruption. This new theme expands and complements previous research, insofar as transparency is understood, in this context, as a concept opposed to that of corruption. This research project will benefit from the international scientific network of its leaders.
Reference
ANR-17-FRAL-0013-01
Website
Programme
ANR - Generic
Type of project
Collaborative research project
Duration
5 years and 5 months
Scientific manager
Guilhem Boulay
Laboratory(ies)
UMR 7300 ESPACE – Study of Structures, Adaptation Processes and Changes in Space
Project coordination
Partners
TELEMME - CDED - ESPACE - LAB'URBA - LATTS
Summary
InveST analyses the implementation of the sustainable development injunction in territorial public action in a context of increased financial rigour. Two hypotheses will be examined. The first states that the rationales, instruments and practices associated with financial stringency influence the selection of priorities and the content of territorial public action in terms of sustainability; and the second that territorial public action under financial constraint contributes to the accentuation of socio-spatial disparities between and within territories. The project will test these hypotheses with a view to overcoming the scientific barrier posed by the literature's failure to take sufficient account of the political and economic constraints and opportunities that underpin policies for the sustainability of territorial systems. Social science research on the role of local authorities in sustainability policies and on the consequences of financial stringency on territorial public action remain largely compartmentalised, making it impossible to analyse the links between sustainability and financial stringency. To remove this barrier, we will focus on the financial practices and power relations that run through the sustainability process. This will enable us to qualify the role of rigour in the transformation of political agendas, objectives and conditions of production of public action for sustainability. One of the original features of the project will be to take account of the diversity of territorial configurations and the specific features of the various areas of public action under study.
Reference
ANR-18-CE22-0004-03
Website
Programme
ANR- Human and Social Sciences
Sub-programme
Arts, languages, literature, philosophy
Type of project
Collaborative research project
Duration
4 years
Scientific Manager
Florence BISTAGNE
Laboratory
UPR 4277 ICTT - Cultural Identity, Texts and Theatricality Laboratory
Project coordination
Avignon University
Partners
Université Lyon II, Università di Napoli Federico II, Université de Lausanne, Università della Campania Luigi Vanvitelli, Universitat de València, Università di Napoli l'Orientale, Università di Napoli Federico II, Museo Nacional de Ceràmica y Artes Suntuarias, Museo Nacional del Prado (Madrid), CESR UMR, MSH Lyon Saint-Etienne,
Summary
The aim of the LUXART programme is to renew the study of the practices and theories of luxury throughout the 15th century in the western Mediterranean area, where the Aragonese monarchy of Naples (1442-1501) exerted its influence. This issue, which is fundamental to Italian culture, must be approached from the perspective of literary, linguistic and artistic studies. On the one hand, the texts on the use of money and the aesthetics of luxury, particularly those by Giovanni Pontano (1429-1503), will be brought out of their theoretical isolation, by means of a digital critical edition with new French translations, and on the other, they will be studied in symbiosis with the works and objects they mention, by placing in this context the works of an artistic heritage that has now been dismembered (manuscripts, medals, paintings, sculptures, domestic ornaments). From Latin to the Neapolitan vernacular, via the Castilian used in the acts of Alfonso V's chancellery, the texts under consideration will make it possible to establish new critical genealogies and re-evaluate an under-exploited corpus on artistic theory and conspicuous consumption in the Renaissance.
Reference
ANR-23-CE54-0002-01
Programme
ANR - Generic
Type of project
PRCE
Duration
3 years and 6 months
Scientific manager
Rachid ELAZOUZI
Laboratory(ies)
UPR 4128 LIA Laboratoire Informatique d'Avignon
Project coordination
Orange
Partners
Nokia Bell Labs France - Inria Grenoble Rhône-Alpes Research Centre - Télécom SudParis - Signals and Systems Laboratory - Centre for Studies and Research in Computer Science and Communications
Summary
By meeting the needs of vertical industries and smart cities, 5G networks will have the capacity to revolutionise our daily lives and our industry. These networks will have to serve traffic with very different requirements, ranging from massive sensor connectivity to real-time remote operation of robots. The concept of slicing makes it possible to satisfy these diverse requirements on a single infrastructure. Slicing makes it possible to create logically isolated network partitions, each slice representing a programmable resource unit (network function, calculation, storage). Slicing was initially proposed for core networks, but there is now talk of using it in Radio Access Networks (RANs) thanks to the development of technologies that enable its implementation: mainly the virtualisation of RAN equipment and the programmability of its control, the advent of Mobile Edge Computing and the flexible design of 5G networks on the physical and MAC layers. However, a number of challenges still need to be overcome before slicing can be fully implemented in the mobile access network, particularly with regard to the management of slices and the associated control and data plans, as well as scheduling and resource allocation mechanisms.
The MAESTRO-5G project is developing technologies for implementing and managing the slices of 5G radio access networks, to enable heterogeneous services to be provided, as well as dynamic infrastructure sharing between operators. To achieve this, the project brings together experts in the fields of performance evaluation, queue theory, game theory and operational research.
The Maestro-5G project must provide :
- a resource allocation framework for slices, incorporating various quality of service constraints and the use of heterogeneous resources;
- a complete slice management architecture that includes provisioning and re-optimisation modules as well as integration into the SDN and NFV layers;
- a business layer for slicing in 5G networks, to ensure that 5G services are commercially viable and accepted by the market;
- a demonstrator, showing the practical feasibility and integration of the main functions and mechanisms proposed in the project, on a 5G access network platform.
This enhanced platform should make it possible to simulate several 5G services and demonstrate the key aspects of slicing, such as:
- the ability to create and operate several slices in parallel, on the same physical infrastructure and sharing the same radio resources, with each slice having its own quality of service constraints,
- the possibility of creating and operating several slices, independent of each other, in parallel, sharing the same infrastructure and belonging to different players, for example different operators;
- demonstrate control between slices, guaranteeing compliance with service contracts and equitable sharing of resources.
Reference
ANR-18-CE25-0012-04
Website
https://anr.fr/Projet-ANR-18-CE25-0012
Programme
ANR-Specific Mathematics in Artificial Intelligence (TSIA) 2023
Sub-programme
Giga-models for automatic processing of natural language and multimodal data
Type of project
Collaborative research project
Duration
4 years
Scientific manager
Mickaël ROUVIER
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
University of Nantes
Partners
University of Aix-Marseille
Nantes University Hospital (CHU Nantes)
Official summary
The recent arrival of Large Language Models (LLMs) and their associated tools for the large
The public consultation process suggests that there are major issues at stake for society. Among the many areas that are, or will be, impacted by
the biomedical field is one of those that is currently attracting the most attention from industry, academia and government.
researchers, but also the general public. Indeed, the need for tools and potential applications seems immense, whether it's for
for example, in the processing of text documents, medical imaging or speech interaction. From
by the sensitive nature of the personal data handled and society's fears associated with tools to help
decision making, work in automatic language processing (ALP) needs to innovate by taking into account the following issues
inherent to this field. As part of the MALADES project, we are proposing innovative approaches to the integration of
LLMs in health centres. The aim is to equip these centres with NLP tools derived from LLMs and adapted to the field of health.
biomedical models and complete control over their health data. The work we
are focusing on four areas of research: 1) the study of the legal and ethical aspects in France of LLMs for the
biomedical field, 2) the integration of voice interaction of LLMs by means of end-to-end approaches, including the collection
speech data, 3) The collection of new original case studies oriented towards the evaluation of models of
generative language, and 4) the integration of dynamic and sovereign LLMs for the biomedical domain, deployed on
and incorporating original approaches that provide LLMs with additional capabilities
by means of knowledge bases that have been mastered and verified.
Reference
ANR-23-IAS1-0005-02
Website
Programme
ANR - Human and social sciences
Sub-programme
Studies of the past, heritage and culture
Type of project
Collaborative research project
Duration
4 years
Scientific Manager
Gerald CULIOLI
Laboratory
UMR 7263 IMBE - Mediterranean Institute of Biodiversity and Ecology EECAR
Project coordination
CNRS
Partners
Avignon Université - University of Montpellier - Archaeozoology, archaeobotany, societies, practices and environments (AASPE)
Summary
This project proposes to bring together archaeologists, archaeometers, traceologists and archaeozoologists to determine the areas of origin of these prestigious objects and to understand the nature and modalities of the relationship between the Maya regions and Costa Rica. We are reviewing all the archaeological data available on the objects under study in order to document their contexts and chronological frameworks. The study of their provenance involves chemical analyses of the stones and adhesives using a range of spectroscopic, chromatographic, dating and isotopic techniques. The micro-traces of the stones' manufacture will be characterised by tracerology in order to understand lapidary techniques and determine whether they were imported or worked locally. The spondylus materials found in the Maya region will be subjected to malacological and isotopic analyses in order to distinguish their geographical sources. The iconographic and epigraphic study of the corpus of plates and mirrors will document their origin, date and status in the Maya area. Finally, all the data will be cross-referenced by being placed in geographical and chronological contexts in order to understand the distribution, production, reuse and functions of these goods in the two regions. This project will enable us to rethink the old distinctions between what is known as Mesoamerica and what is often considered an 'intermediate zone', in order to define the entities that make them up in a much more dynamic way.
Reference
ANR-22-CE27-0023-05
Website
Programme
ANR - Digital sciences
Sub-programme
Interaction, robotics
Type of project
Company collaborative research project
Duration
4 years
Scientific Manager
Fabrice LEFEVRE
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
Avignon University
Partners
Hubert Curien Laboratory (Jean Monnet University) - INRIA Grenoble Research Centre - APHP Alzheimer's disease
Summary
In muDialBot, our ambition is to proactively incorporate human behavioural traits into human-robot spoken communication. We plan to reach a new stage in the exploitation of the rich information provided by audio and visual data streams from humans. In particular, extracting verbal and non-verbal events should make it possible to increase the decision-making capabilities of robots so that they can manage speaking turns more naturally and also switch from group interactions to face-to-face dialogues depending on the situation.
Recently, there has been growing interest in companion robots capable of assisting individuals in their daily lives and communicating effectively with them. These robots are perceived as social entities and their relevance to health and psychological well-being has been highlighted in studies. Patients, their families and healthcare professionals will be able to better appreciate the potential of these robots, as certain limitations are rapidly overcome, such as their ability to move, see and listen in order to communicate naturally with humans, beyond what is already possible with touch screens and voice commands.
The scientific and technological results of the project will be implemented on a commercial social robot and will be tested and validated with several use cases in the context of a day hospital unit. Large-scale data collection will complement the in-situ tests to fuel future research.
Reference
ANR-20-CE33-0008
Website
Programme
ANR - Digital sciences
Sub-programme
Characterisation of the structures and structure-function relationships of biological macromolecules
Type of project
International collaborative research project
Duration
3 years and 8 months
Scientific Manager
Grégory DURAND
Laboratory
UPRI - ERIT S2CB
Project coordination
Avignon University
Partners
Leipzig University (Germany) - Technische Universität Kaisersalutern (Germany)
Summary
Membrane proteins (MPs) play a fundamental role in cell function, controlling communication between cells and with their environment, as well as the transport of nutrients. PMs currently account for around half of all therapeutic targets. Unfortunately, their characterisation remains tricky because of the need to maintain an environment that mimics the native membrane. Conventional methods for isolating PMs often rely on inappropriate and aggressive chemistry, leading to their denaturation.
Among the alternative chemical strategies, phospholipid nanodiscs and amphiphilic styrene/maleic acid copolymers (SMA) have great potential. SMA can form patches directly from natural or artificial membranes. These 'patches', consisting of a double lipid layer into which the protein is inserted, mimic the lamellar organisation of cell membranes. However, this promising technology is limited by the fact that the polymers currently available affect the dynamics of the lipids and the protein, which is an essential element in the function of PMs. Two other limitations are the presence of aromatic rings, which absorb strongly under UV light, and the high charge density due to the presence of carboxyl groups on the polymer chain. In this context, NanoBelt aims to develop new amphiphilic polymers combining a high capacity to solubilise membranes while having a moderate impact on lipid dynamics, low UV absorption, negligible charge density, and the ability to maintain the function and conformational dynamics of encapsulated proteins.
NanoBelt is based on the observation that the solubilising properties of polymers are highly dependent on several structural parameters (nature and distribution of monomers along the polymer, hydrophobicity, degree of polymerisation, etc.) which in turn modulate the conformational dynamics of proteins and lipids. These crucial variables cannot be identified or improved unless series of polymers with well-defined properties are produced.
Reference
ANR-20-CE92-028
Website
Programme
ANR - Digital sciences
Sub-programme
Foundations of digital technology: computer science, automatic control, signal processing
Type of project
Collaborative research project
Duration
4 years
Scientific Manager
Yezekael HAYEL
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
CNRS
Partners
Centre de recherche en automatique de Nancy (Université de Lorraine) - Laboratoire des signaux et systèmes (Université Paris-Saclay) - - Bureau d'Economie Théorique et Appliquée (Université Strasbourg Lorraine)
Summary
In the NICETWEET project we are developing an innovative, comprehensive and unified methodology to address an important problem that arises in many areas, such as economics, finance and politics. Several decision-makers find themselves competing to promote their products, services or ideas within a large group of agents who are connected via a physical or digital social network. The agents are therefore under the (endogenous) influence of their neighbours in the social network graph, but also under the exogenous influence of the decision-makers. The latter have some knowledge of the social network and the underlying opinion dynamics, and use this to implement targeted influence mechanisms. To tackle this intrinsically interdisciplinary problem, we have assembled a team of researchers with expertise in the following areas: control theory and, more specifically, opinion dynamics, game theory, information theory, complex networks and economics. In contrast to the closest work, we assume that the social network has certain characteristics that are essential for realistically modelling the economic applications of interest that are being addressed. Very importantly, we also assume the presence of multiple decision-makers. Our preliminary knowledge of economic applications indicates that some key features need to be taken into account simultaneously. These characteristics are: the social network is very large and parsimonious; agents can join or leave the network at any time; decision-makers may have imperfect knowledge of the social network parameters and opinion dynamics; some agents may exchange not only their opinion but also their reliability; possible presence of extremists. A fundamental characteristic with regard to the state of the art is that the dynamics of opinion can be controlled, and moreover by several decision-makers. To deal with this problem of controlled opinion dynamics in the presence of decision-makers who may have divergent interests (i.e. non-aligned utility functions), we will make use of game theory and help to build bridges between the formal control literature and the much less formal economics literature. In particular, one of our technical objectives will be to develop a formal and systematic technique for designing viral marketing strategies that are both effective and implementable. An important issue in the design of viral marketing strategies is the policy of spatio-temporal allocation of a marketing, advertising or influence budget, the problem being how to distribute a budget between social network agents and over time. Effectiveness will be measured in terms of the quality of exploitation of the knowledge available to a decision-maker, as well as in terms of strategic reaction. To design these strategies and manage the risk specific to certain research avenues, we will adopt several approaches in parallel. One approach will be to exploit theorems for characterising boundary performance under partial information for stochastic (repeated) game models as a design guide. Another will be to exploit and adapt multi-player learning rules such as those derived from Bayesian multi-agent learning. To implement NICETWEET's ambitious roadmap and carry out the corresponding research, the recent and promising results obtained within the consortium will be exploited as a basis for the project.
Reference
ANR-20-CE48-009
Website
Programme
ANR - Life sciences
Sub-programme
Physiology and pathophysiology
Type of project
Collaborative research project
Duration
4 years
Scientific Manager
Cyril REBOUL
Laboratory
UPR 4278 LAPEC - Laboratory of Cardiovascular Experimental Physiology
Project coordination
Avignon University
Partners
Physiology and experimental medicine of the heart and muscles - Medicines and Health Technologies
Summary
Increased cardiac mechanical stress leads to functional and possibly structural adaptations of the myocardium when the mechanical stress persists. Repetition of these mechanical stresses can lead to physiological remodelling of the heart, as observed in response to physical exercise. However, when this stress becomes chronic, it can lead to unfavourable remodelling associated with functional inconsistencies and thus to the development of cardiac pathology. To date, the main treatment for limiting the onset of deleterious cardiac remodelling is to reduce pressure overload. It therefore seems essential to improve our understanding of the signalling pathways involved in the functional and structural response of the heart to increased ventricular load stress.
In response to stretch, the myocardium increases its production of nitric oxide (NO). Until recently, NO signalling was thought to be linked to activation of the cyclic guanosine monophosphate (cGMP)-dependent protein kinase (cGMP-NO) pathway. However, NO can also modulate cell signalling by binding directly and covalently to protein cysteine residues, a process known as S-nitrosylation (SNO). This second mechanism has been less studied and is the focus of the NitrosoCard project. SNO is a major reversible post-translational modification involved in regulating the activity of proteins and also protecting them from irreversible oxidation. This process has been well described in mitochondria, where it is associated with reduced production of reactive oxygen species and cardio-protection. Cardiac myofilaments are well described as being sensitive to redox signalling phenomena and are therefore potentially sensitive to SNO. However, to date the role of SNO in the regulation of the cardiac contractile machinery remains unclear.
In the NitrosoCard project, we hypothesise that cardiac stretch stress causes translocation of endothelial NO synthetase (eNOS)-SNO signalling to myofilaments, thereby regulating their contractile properties. When myocardial stretch stress becomes too severe and chronic, this cell signalling becomes defective and could help to explain the longer-term development of a pathological response in the heart.
The objectives of the NitrosoCard project are to elucidate the role of SNO in sarcomeric proteins in response to stretch stress in physiological or pathological conditions. Our hypothesis will be tested through three original objectives that will enable us to approach the whole animal right down to the molecular level. The NitrosoCard project will aim to assess 1- the impact of myocardial stretch stress on the translocation of eNOS/NO/SNO signalling to myofilaments; 2- the direct or indirect impact of SNO on the function of the contractile machinery in response to myocardial stretch; and finally 3- whether the repetition (physical activity) or chronicity (pathology) of myocardial stretch stress leads to different S-nitrosylation profiles of contractile machinery proteins and thus to different adaptive response profiles in the heart.
The NitrosoCard project should make it possible to identify new redox-dependent signalling pathways involved in the physiological response of the heart to stress. The loss of this NO-dependent signalling could play a key role in the transition to pathological remodelling of the heart. The identification of these mechanisms could lead to the identification of new molecular targets that are crucial for the development of new therapeutic strategies.
Reference
ANR-21-CE14-0058-01
Website
Programme
JPI: Cultural heritage, society and ethics
Sub-programme
Type of project
JPI
Duration
4 years
Scientific Manager
Julie DERAMOND
Laboratory
UMR 8562 CNE - Dynamics of Social Worlds
Project coordination
Avignon University
Partners
Politecnico de Leiria (Portugal) - University of Thessaly (Greece)
Summary
OLIVE4ALL is based on the idea of raising awareness of sustainable development through heritage. Our aim is to develop an innovative, interdisciplinary and transferable methodology, based on the study of heritage linked to the olive and olive tree, chosen because it is representative of Mediterranean identity and conveys strong values and symbols that are not sufficiently promoted in the Euro-Mediterranean region. Based on a critical approach to heritage, we want to study and make visible a rural heritage that is often neglected, as well as the players and communities linked to it, not all of whom are aware of the societal value of this heritage. OLIVE4ALL retraces and highlights the process of establishing heritage and heritage communities around the olive tree, questioning the very concept of heritage. The project proposes an original experiment centred on a transmedia device that studies how carefully developed forms of mediation based on narrative, sensory and digital elements can be used to raise the awareness of players at several territorial levels or audiences far removed from heritage. OLIVE4ALL is therefore taking a close look at ways of providing inclusive access to heritage. Once the idea has been grasped, the next step is to take action. We will be developing specific tools to help local players to talk to each other and network, so as to integrate heritage into the development of their area. OLIVE4ALL's aim is to raise the profile of heritage in society, through actions designed to strengthen public policies. OLIVE4ALL's aim is to demonstrate that by giving heritage a more prominent place, we can respond to the global challenges that the current health crisis is making even more urgent. Enhancing the value of food heritage in European societies raises awareness of the human, social, economic and environmental dimensions of sustainable development. To ensure that OLIVE4ALL has the widest possible impact, we are mobilising a wide range of stakeholders who are invited to participate and collaborate, in order to share knowledge, learn from best practice and bring about lasting change, to build the resilient society of tomorrow.
Reference
ANR-21-CHIP-0002
Website
Programme
ANR - Generic
Type of project
PRCE
Duration
4 years and 3 months
Scientific Manager
Yannick ESTEVE
Laboratory(ies)
UPR 4128 LIA Laboratoire Informatique d'Avignon
Project coordination
Avignon University
Partners
Laboratoire d'informatique de l'Université du Mans - Université Grenoble Alpes - Airbus Defence and Space SAS
Summary
The ON-TRAC project proposes to radically change the architectures currently used in speech translation. It is based on end-to-end neural models for automatic translation and is aimed more specifically at the lightweight, portable speech translation applications that Airbus is developing for security operations in theatres of operation.
In addition to the study of end-to-end approaches based on language pairs associated with large-scale training data, ON-TRAC will study the development of models for poorly endowed oral or dialectal languages.
An end-to-end approach to speech translation, as envisaged here, would make it possible to review the data collection methodology for the development of a speech translation system.
With this approach, there is no need to transcribe the source language: the cost of producing the data needed to learn a speech translation system is therefore greatly reduced, and the development of such a system for new languages (including those without a writing system) would be facilitated and accelerated.
Since the project is aimed at portable translation applications, ON-TRAC is also interested in studying the computing time and memory footprint required for neural speech translation.
ON-TRAC will enable the processing of three distinct language pairs with increasing security and defence operational interest and level of difficulty (English-French; Pashto-French; Tamacheq-French).
The ON-TRAC project is part of Axis 4 "Data, Knowledge, Big Data, Multimedia Content, Artificial Intelligence" of Challenge 7 "Information and Communication Society" of the ANR's 2018 action plan.
With its main scientific theme dedicated to speech translation using end-to-end neural approaches, it is clearly positioned in the "From data to knowledge" and "Multimedia content processing" themes.
The technologies developed in the ON-TRAC project will be tested on three language pairs, with written French as the systematic target language.
The first pair of languages studied will be spoken English into written French, for reasons of simplicity and for a better perception of the phenomena that occur during translation through the analysis of the output of our systems, English being sufficiently mastered by all the players in the project.
The Pashto language will be the source language for the second language pair. This choice is dictated by the fact that the treatment of an oral dialect is one of the stated objectives of the project, and by the fact that the cost of collection will be kept to a minimum, since the consortium already has around a hundred hours of audio recordings in Pashto, with their textual translations into French (as well as their transcription into Pashto).
Finally, the third language pair will be Tamacheq, an oral dialect spoken by the Tuaregs in various areas of interest for intelligence and security (Sahel, Niger, Mali, Burkina Faso, Libya, etc.). As such, it is of great interest to the government services concerned.
Reference
ANR-18-CE23-0021-01
Website
Programme
ANR-Specific Mathematics in Artificial Intelligence (TSIA) 2023
Sub-programme
Giga-models for automatic processing of natural language and multimodal data
Type of project
Collaborative research project
Duration
3 years
Scientific manager
Yannick Estève
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
Grenoble Alpes University
Partners
National Audiovisual Institute (INA)
UNIVERSITE PARIS CITE
Centre National de la Recherche Scientifique (CNRS) - Ile-de-France Regional Office Gif-sur-Yvette
Official summary
The Pantagruel project is an ambitious initiative to develop and evaluate multimodal (written, spoken, pictograms) and inclusive linguistic models for French. The project draws on the expertise of researchers from a range of disciplines, including computer science, signal processing, sociology and linguistics, to ensure diversity of perspective and reliability and relevance of results. The main contributions of the project are the development of models
freely accessible self-supervised models for French, comprising one to three of the modalities for general and clinical domains. In addition to producing models, the project will also design test beds to assess the generalisation of this type of model, drawing on the experience gained from the FlauBERT and LeBenchmark projects. Part of the project will be devoted to the biases and stereotypes conveyed in the training corpora and in the downstream models. We will be working with an ethics committee to limit the amplifying effect of bias within the training corpora, in particular by working on the demographic characteristics of speakers (for oral audio or transcriptions) and authors (for some of the written data). This will enable us to compare the models learned on training corpora with varying proportions of these characteristics, such as gender. This study will make it possible to quantify the extent to which the predictions of the language models are reliable reflections of the upstream corpora, and to better control the way in which they can be used as research tools for the social sciences. The project will develop software components that will facilitate the integration of language models into various applications and enable the development of innovative solutions that exploit the power of multimodal French language models. These tools are intended in particular for non-computer scientists such as the members of the consortium (sociologists, linguists, doctors, speech therapists), researchers in other fields and artists. The Pantagruel project thus has the potential to significantly advance the state of the art in multimodal language models and to have a positive impact on a wide range of applied fields, from healthcare to the arts and the humanities and social sciences.
Reference
ANR-23-IAS1-0001-04
Website
Programme
ANR - Digital sciences
Sub-programme
Multi-usage communication networks, high-performance infrastructures, software science and technology
Type of project
Collaborative research project
Duration
3 years and a half
Scientific Manager
Francesco DE PELLEGRINI
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
Avignon University
Partners
Centre d'études et de recherche en informatique et communications (CEDRIC) - Inria Sophia Antipolis-Méditerranée Research Centre - Parallelism Computing Laboratory - Computer Science, Systems, Information and Knowledge Processing Laboratory
Summary
The new generations of mobile access networks promise high-speed communication, reduced delays and processing capabilities offered by the network itself. Data generated by the IoT (video, for example) or by users' smartphones will feed AI applications running on edge/fog servers. PARFAIT tackles the new problems of resource allocation for AI applications made up of contained microservices. This allocation poses major problems because it now has to be orchestrated in a distributed way, while the footprint of modern AI applications with learning capabilities remains unknown. The PARFAIT project is developing theoretical foundations for the efficient allocation of distributed and scalable resources on state-of-the-art IT infrastructures carrying out AI processing. Algorithmic solutions will be developed based on the theory of constrained and distributed Markov decision processes to orchestrate service placement and resource allocation and to quantify the effect of orchestration policies. In addition, by formalising our problem through the prism of team game theory, the project will pave the way for decentralised orchestration, a missing element in meeting the need for data and application proximity and the synchronisation problems that arise when several orchestrators cooperate by making decisions based on a local or partial view of the system. In addition, in order to achieve efficient and on-the-fly orchestration of services (edge), these solutions will be equipped with reinforcement learning techniques in order to define a set of orchestration algorithms capable of adapting over time to the load of applications and dealing with uncertain information regarding the footprints of AI applications. Validation will be carried out with a view to demonstrating concrete solutions for practical orchestration use cases, taking advantage of large-scale simulation and research testbeds.
Reference
ANR-21-CE25-0013-01
Website
Programme
ANR - Generic
Type of project
Company collaborative research project
Duration
5 years and 3 months
Scientific Manager
Driss MATROUF
Laboratory(ies)
UPR 4128 LIA Laboratoire Informatique d'Avignon
Project coordination
Avignon University
Partners
A.I.MERGENCE - INRIA Nancy Grand Est
Summary
During periods of inactivity, using an autonomous mobile robot to monitor industrial premises is an extremely cost-effective solution. The robot moves around the premises and analyses activity. When a person is detected, the robot is tasked with checking their identity. If there are any problems, the robot contacts a human operator. One of the major objectives of this project is to take into account as realistically as possible the real conditions in which the robot is used. This means conducting experiments on the robot itself and in a realistic environment. Voice identification in the context of a mobile security robot faces a number of challenges relating to the remote identification of a person in real-life conditions, which can currently drastically reduce performance: ambient noise and the robot's internal noise (linked to the robot's activators) which have an impact on the audio sensors, resulting in low signal-to-noise ratios (SNRs), reverberation phenomena due to the configuration of the highly variable locations in which the robot is located, the variable location of the speakers, etc. In this project, we are proposing methods and approaches to try and overcome the various scientific obstacles mentioned above. The proposed solutions are based on our expertise in acoustic modelling and signal processing, as well as on the use of deep neural networks. Deep neural networks are at the heart of machine learning research in a number of fields, and go beyond the purely statistical methods used until now.
Despite efforts to remove acoustic barriers, there are scenarios in which voice identification alone will not offer total reliability. In applications where a high level of security is required, the use of a single modality is often too risky and voice identification is often implemented in conjunction with other identification modalities. To meet this need, the robot proposed in this project uses its ability to interact with the people detected. This modality is used when the robot does not have enough information to make a reliable decision. It can use its interaction capabilities to acquire more acoustic data in order to consolidate voice authentication. The robot can also use the interaction module to remove ambiguity through a set of simple questions and answers based on knowledge that can be verified by the robot (for example, asking for the first name or surname of the direct supervisor of the person being inspected). Finally, information about the speaker's emotional state and the acoustic scene will be transmitted to the system so that it can adapt the dialogue strategy, the robot's behaviour and the pre-processing and voice identification algorithms.
In addition to the direct scientific and technical expectations, this project will provide an opportunity to create and disseminate a unique body of work that will make it possible, during and after the project, to evaluate the solutions developed to overcome various obstacles, such as ambient noise, reverberation and short duration. An evaluation plan with an experimental protocol will be defined to ensure that the solutions developed during the project are relevant for both the scientific community and the industrial partner.
Reference
ANR-18-CE33-0014-01
Website
Programme
Horizon 2020
Sub-programme
RIA (Research and Innovaction Actions)
Duration
2 years and 3 months
Scientific manager
Yannick ESTEVE
Laboratory(ies)
UPR 4128 LIA Laboratoire Informatique d'Avignon
Project coordination
Partners
Latvijas Universitates Matematikas Un Informatikas Instituts (LV) - Priberam Informatica SA (PT) - Fraunhofer Gesellschaft Zur Foerderung Der Angewandten Forschung EV (DE)
Summary
Large amounts of multilingual text and speech data are available on the internet, but until very recently the potential for exploiting this data has remained untapped. Recent advances in deep learning and transfer learning have opened up new possibilities.
opening the door to new possibilities - in particular, the integration of knowledge from these large sets of unannotated data into plug-in models to tackle machine learning tasks.
The aim of Stream Learning for Multilingual Knowledge Transfer (SELMA) is to address three tasks: to ingest large amounts of data and continuously train machine learning for several natural language tasks; to control these data flows using these models in order to improve the monitoring of multilingual media; and to improve the production of content for multilingual media;
production of multilingual media content, closing the loop between monitoring and content production.
SELMA has eight objectives:
1. Enable the processing of massive streams of video and text data in a distributed and scalable way.
2. Develop new methods for training unsupervised deep learning linguistic models in 30 languages.
3. Enable the transfer of knowledge between tasks and languages, by supporting low-resource languages.
4. Develop new methods of data analysis and visualisation to facilitate the decision-making process in media monitoring. 5. To develop an open-source platform to optimise the production of multilingual content in 30 languages.
6. Refine deep learning models based on user feedback, reducing the repetition of errors.
7. Ensuring the sustainable operation of the SELMA platform
8. Raise awareness and encourage the active involvement of users in the platform.
To achieve these objectives, we need to advance the state of the art in a wide range of technologies: transfer learning, linguistic modelling, speech recognition, automatic translation, summaries, etc,
speech recognition, machine translation, summarisation, text-to-speech, named entity association, learning from user feedback.
Reference
957017
Website
Programme
ANR - Generic
Type of project
PRC
Duration
4 years
Scientific Manager
Corinne FREDOUILLE
Laboratory(ies)
UPR 4128 LIA Laboratoire Informatique d'Avignon
Project coordination
Toulouse 3 University
Partners
CHU Toulouse Research Department - Speech and Language Laboratory - Octogone Interdisciplinary Research Unit
Summary
In the context of speech production disorders observed in ENT cancers and neurological, sensory or structural pathologies, the aim of the RUGBI project is to improve the measurement of intelligibility deficit. Speech production disorders can lead to a serious loss of intelligibility, making it difficult for patients to communicate with those around them and limiting their professional and/or social lives. Traditionally, clinical assessment of intelligibility has been based on global perceptual assessment, which is considered unsatisfactory due to its subjectivity, lack of precision and duration, leading to erroneous measurements of patients' intelligibility. In addition, the speech production tasks used for this type of assessment (repetition of words, sentences, reading) are far from suitable for accurate measurement of intelligibility and only allow an overall assessment of functional impairment.
RUGBI aims to overcome these limitations by developing a new objective assessment tool based on i) the identification of relevant linguistic units, from an acoustic and prosodic point of view, and ii) the identification of sensitive linguistic tasks. The aim of the RUGBI project is thus to supplement the therapist's tools with a precise, robust and rapid measure that will enable an optimised therapeutic plan to be developed with a view to achieving a tangible improvement in intelligibility.
To achieve this, RUGBI is relying on large corpora, already available, containing speech productions from healthy subjects (190) and patients (365) with pathologies of structural origin (VADS cancers) and neurological origin (Parkinson's disease), during the performance of different linguistic tasks, and for some of them, at different stages of the disease. These corpora are a considerable asset for conducting the project's two lines of study, based respectively on i) the perception of speech intelligibility and ii) modelling by automatic speech processing, and more particularly, on Deep Learning and its data representation properties, which will be exploited here. In this context, the central objective of the project brings together the expertise of its members from the medical field, the language sciences and speech and language engineering to meet the challenges of biology and health. RUGBI's multidisciplinary expertise is a guarantee of success.
Reference
ANR-18-CE45-0008-04
Website
Programme
ANR - Energy and materials sciences
Sub-programme
Polymers, composites, soft matter physics and chemistry
Type of project
Collaborative research project
Duration
4 years and 3 months
Scientific Manager
Raphaël PLASSON
Laboratory
UMR_A 408 SQPOV - Safety and Quality of Plant Products
Project coordination
Sorbonne University
Partners
Institut Charles Sadron (University of Strasbourg) - Institut des sciences chimiques de Rennes (University of Rennes I) - Selective Activation Process by Uni-electronic or Radiative Energy Transfer (ENS)
Summary
To endow molecules and materials with properties, chemists can rely on established structure-property relationships, which is relevant at the molecular and supramolecular levels where a great deal of information exists for the synthesis and prediction of physico-chemical properties, but remains tricky for designing complex functions integrating multiple interactions and reactions. A reverse property-structure strategy can then be adopted by screening libraries of interacting/reacting constituents for target properties. However, screening is time-consuming, limiting the size of the libraries and therefore the diversity of the system. Biologists ignore this limitation. They condition the survival of living cells on the property they want to obtain. This powerful strategy is currently unprecedented in an abiotic system, which is what motivated the SACERDOTAL project. The aim is to introduce and validate a selection/survival protocol in a monomer/polymer chemical system for directed evolution towards a property.
Our protocol applies a temperature gradient to monomers engaged in reversible ligation/hydrolysis reactions. This out-of-equilibrium state continuously forces the system to randomly explore possible polymer lengths and sequences. Among the polymers formed, the selection and survival of a sub-population is conditional on achieving the desired property by applying experimental constraints. We have adopted amino acids and peptides as monomers and polymers, the latter being widely used to produce water-soluble nanoparticles for biomedical applications. We therefore propose to select and produce peptide-conjugated nanoparticles using a survival assay that selects compounds that bind to nanoparticle surfaces and generate luminescent conjugates. This is an important objective for imaging, as these nanoparticles still suffer from drawbacks (toxicity, lack of brightness and/or colloidal stability), which could be overcome with optimised protective shells for the inorganic core.
This integrative, multidisciplinary project includes original chemical developments (dynamic combinatorial amino acid libraries, nanoparticle synthesis protocols), theoretical and numerical simulations of out-of-equilibrium polymerisation (from reaction-diffusion models to stochastic metadynamic models), the design and production of instrumental devices (microsystems, optical heating and fluorescence imaging) and analyses of nanoparticles and their conjugated peptides using established or innovative techniques.
In addition to the production of functionalised nanoparticles, the validation of our selection/survival protocol will also be significant from a methodological perspective, and it will thus be patented for technology transfer and economic development. This protocol addresses theoretical, instrumental and analytical aspects that should be common to all work in this emerging field. It uses discriminating criteria (diffusion coefficients and reaction rate constants) that go beyond the thermodynamic criteria used in current screening protocols. It is compatible with numerous chemistries and has a molecular diversity greater than that currently available to micro-organisms, free from any physiological or toxicity constraints. In this project, it exploits peptides whose potential for interactions and catalysis provides an attractive platform for screening numerous properties of Darwinian chemical systems (self-assembly, non-equilibrium dynamics, catalyst discovery).
Reference
ANR-19-CE06-0010-02
Website
Programme
Sub-programme
Type of project
Young researchers
Duration
4 years
Scientific Manager
SANJUAN Eric
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
University of Brest
Partners
University of Minho (Portugal) - Université d'Avignon - Université Bretagne Sud - Université d'Aix-Marseille - University of Amsterdam (Netherlands)
Summary
Information systems give citizens access to information from reliable sources such as scientific literature. The latter is difficult for non-experts to access because of the complexity of the language and the lack of scientific acculturation. Individual or political decisions can be affected by a lack of knowledge of scientific work, as was the case during the COVID-19 pandemic. The automatic simplification of texts aims to reduce these obstacles and could be useful in various fields such as scientific communication and journalism, politics and education. SimpleText will be a step forward in making research truly accessible and understandable to all, and will help to counter fake news based on scientific results (ODD QUALITY EDUCATION). Simplified texts are also more accessible to non-native speakers, young readers and people with reading disabilities (REDUCED INEQUALITY SDG). SimpleText aims to answer the following research questions: RQ1 - What information-carrying textual expressions need to be simplified (documents and passages to be included in the simplified summary)? RQ2 - What types of basic information should be provided (which concepts should be contextualised by giving a definition, an application, an example, etc.)? RQ3 - How can the readability of a given short text be improved (e.g. by reducing vocabulary and syntactic complexity) with an acceptable rate of alteration of the information? RQ4 - To what extent are the approaches used for English applicable to French? Answering these questions presupposes providing algorithms, data and measurements that will be used in international evaluation campaigns (e.g. CLEF), as well as during courses in digital humanities, website localisation, pre-editing and technical writing.
Reference
ANR-22-CE23-0019
Website
Programme
ANR - Setting up European or international scientific networks
Sub-programme
Type of project
Setting up European or international scientific networks
Duration
2 years
Scientific Manager
Christine PEPIN
Laboratory
UPRI
Project coordination
Avignon University
Partners
Bracco Suisse (Switzerland) - University of Zurich (Switzerland) - University of Geneva (Switzerland) - Institute for Cancer Research (United Kingdom) - Sorbonne University (France) - University of Magdeburg (Germany) - SINTEF Industry (Norway)
Summary
The aim of this project is to study the use of monodispersed perfluorocarbon (PFCD) droplets instead of
MB contrast agents currently used in all BBB opening clinical trials. These 'BBB openers' (D1
droplets) will generate tailor-made bubbles with specific physico-chemical characteristics (type of shell, size and shape).
sonosensitivity) will be designed using molecular dynamics as well as sonoporation simulations that will help
in understanding and optimising the relevant processes, in order to achieve a safe and controlled BBB
permeabilisation. One of the added values of the project will be to develop real-time control of the opening of the BBB by
combining optoacoustic tomography and magnetic resonance imaging (MRI) guidance/monitoring. We will develop sophisticated drug carriers (D2 droplets) capable of improving the therapeutic window of the
The process of opening the FUS-BBB by (i) increasing the cerebral concentration of the drug (better diffusion through the
BBB, limited interaction with efflux pumps, concomitant administration of efflux inhibitors), (ii) enabling the
localised release of the drug on demand under non-invasive and specific acoustic activation (drug triggered by the FUS)
version optimised by MRI monitoring). Finally, in vitro and in vivo experiments will validate the therapeutic efficacy of the treatment.
potential of our two families of sono-activated tools (D1 for BBB opening and D2 for drug delivery) on
glioblastoma, the chosen 'model' brain disorder. We will focus on the controlled administration of drugs to the brains of
temozolomide, which to date has been the most effective chemotherapeutic drug for treating glioblastoma, but we're making it
does not rule out the possibility of applying this technology to other molecules, especially if the clinical trials currently underway show that this is the case.
their merits in the near future.
Reference
ANR-22-MRS3-0001-01
Programme
ANR - Cross-cutting areas
Sub-programme
Technologies for health
Type of project
Collaborative research project
Duration
4 years
Scientific Manager
Christine PEPIN
Laboratory
UPRI
Project coordination
Sorbonne University
Partners
Avignon University - National Institute for Health and Medical Research (iBrain) - ENS Physics Laboratory
Summary
The central nervous system (CNS) is a therapeutic target for many brain disorders, including Charcot's disease (ALS), which is the subject of our project. Although the blood-brain barrier (BBB) is known to be permeable in certain brain diseases, crossing the BBB remains a major challenge for treating most CNS diseases, including ALS. Our project consists of creating a tool that causes the transient opening of the BBB, followed by the controlled release of an active substance into the CNS. To achieve this, we will use a cocktail of large and small perfluorocarbon drops sensitive to different acoustic sequences. Ultrasonic activation of the large drops will open the BBB, allowing the small drops to pass through intact. Then, another acoustic sequence different from the first will induce the release of the active principle from the small drops in the cerebral compartment.
Reference
ANR-22-CE19-0031-02
Website
Programme
ANR - Environmental sciences
Sub-programme
Food and food systems
Type of project
Collaborative research project
Duration
5 years
Scientific Manager
Guillaume WALTHER
Laboratory
UPR 4278 LAPEC - Laboratory of Cardiovascular Experimental Physiology
Project coordination
Avignon University
Partners
Grenoble Alpes University Hospital - Taste and Food Science Centre - LBTI Tissue Biology and Therapeutic Engineering (Lyon I University) - Inserm
Summary
Consumption of products containing non-nutritive sweeteners is growing rapidly worldwide and appears to be associated with weight gain and the risk of developing chronic pathologies, particularly cardio-metabolic diseases. However, the underlying mechanisms needed to better decipher the consequences for human health are not known. Recent studies have identified the presence of sweet taste receptors (T1R) in the pancreas and intestine, but also, more surprisingly, in the brain and endothelial cells. We therefore hypothesise that these sweetener-activated T1Rs are involved directly (at the level of the vessel) and indirectly (via central integration or metabolic modulation) in vascular reactivity, an early marker of the development of cardiometabolic pathologies.
Reference
ANR-19-CE21-0003-01
Website
Programme
Digital sciences
Sub-programme
Artificial intelligence and data science
Duration
4 years
Type of project
Collaborative research project
Scientific manager
Yannick Estève
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
University of Lorraine
Partners
Urecom
Official summary
As defined by the General Data Protection Regulation (GDPR), voice data falls within the scope of personal data. Its use
in applications is also considered to present a high risk in the future "AI act". This is because speech recordings contain much more than just content
(words), but also, for example, the speaker's identity, gender, age, regional accent, etc. All this personal and private information can potentially be
estimated from voice data and used for malicious or unwanted purposes. Technologies to improve privacy are needed, for users of
voice technologies in order to prevent their voice recordings from being used for purposes for which the user has not given their consent. Protection of privacy
for speech data is still underdeveloped due to a number of challenges. The SpeechPrivacy project envisages an approach that goes far beyond the solutions already available.
by reducing the need to trust voice service providers and giving users total control over their confidentiality. So
the user will be able to choose which (privacy-sensitive) attributes he or she will be able to give the service provider access to. SpeechPrivacy will offer a flexible solution for privacy preservation, based on isolated/unentangled representations and obfuscation of selected attributes. The scenarios of use are multiple: protection of witnesses, preservation of the privacy of teenagers online and in social media, storage of medical data recordings (removing the patient's personal information and preserving the quality of the signal for research access). The objectives of SpeechPrivacy are to propose (1) specific and optimised solutions for disentangling age, gender, vocal identity and regional accents, among other speech attributes; (2) a solution capable of disentangling several attributes, including those mentioned in (1), and assessing the impact of their obfuscation on other attributes as well as the usefulness of the resulting signal on the given use cases; (3) a solution for the robust detection of sensitive words and speaker age/gender information in the linguistic content, in order to replace them in the speech signal; (4) a user interface to give users full control over the privacy/utility trade-off of the resulting signal. The research is organised into several distinct tasks in order to achieve our objectives. First, we will develop solutions for obfuscating specific and unique speech attributes, using a one-versus-all modelling approach. We will then extend this work to the consideration of multiple attributes and their disentanglement, using joint and adversarial learning. We will develop a framework in which distinct attributes are explicitly modelled by specific dimensions in a common representation that will allow multiple selected attributes to be obfuscated simultaneously. The third task concerns semantic privacy and will address the protection of sensitive named entities, age and gender attributes in linguistic content and their obfuscation in the acoustic signal. Finally, the design of databases, protocols and measurements, common to all tasks, will be studied in a fourth task with demonstration activities designed to ensure international dissemination and impact.
Reference
ANR-23-CE23-0022-03
Website
Programme
ANR - Generic
Type of project
PRCE
Duration
4 years
Scientific manager
Jean-François BONASTRE
Laboratory(ies)
UPR 4128 LIA Laboratoire Informatique d'Avignon
Project coordination
Partners
Summary
TheVoice project tackles the creation of voice for content production in the creative industry sector (films, series, documentaries), a sector that is very important in terms of industrial potential but extremely demanding in terms of quality. The project is based on a simple observation: voice production is still carried out exclusively by human operators in an almost exclusively digital sector. The scientific and technological objectives of the project are to model an actor's 'vocal palette' to enable voices to be recommended on the basis of similarity, and to create artificial voices capable of reproducing an actor's vocal identity. The project will create a breakthrough in usage by developing and industrialising new technologies for the creation of natural, expressive voice content. The consortium, led by a major player in the digital content creation industry and made up of recognised research laboratories, aims to consolidate a position of excellence in digital research and technologies "Made-in-France" and to promote French culture throughout the world.
Reference
ANR-17-CE23-0025-02
Website
https://anr.fr/Projet-ANR-17-CE23-0025
Programme
ANR- Specific Support for Defence Research and Innovation (ASTRID)
Sub-programme
Cognitive Warfare
Type of project
Collaborative research project
Duration
3 years
Scientific Manager
Yannick ESTEVE
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
University of Lorraine
Partners
Summary
4th generation warfare (4GW) is known as information warfare involving populations that are not necessarily military. It is waged by national or transnational groups following ideologies based on cultural convictions, religious dogmas, economic or political interests, with the aim of sowing chaos in a targeted part of the world. With the emergence of social networks, the battlefield, whose contours were once blurred, has found a place for the 4GW. Indeed, one of the 4GW's points of attack is the massive use of social networks to manipulate minds. The aim is to prepare public opinion in one part of the world to accept a state of affairs and make it humanly acceptable and politically correct. This can happen prior to the invasion of a country, in the run-up to a presidential election, to provoke a collapse, increase energy prices or destabilise the political situation in countries where the democratic regime is faltering, etc.
One of the modalities of 4GW is what is known as cognitive warfare. Like fourth-generation warfare, the aim is to blur the mechanisms for understanding politics, economics, religion and so on. The consequence of this action is to destabilise and reduce the adversary. This cognitive war therefore targets the brain of what is supposed to be the enemy. Ultimately, 4GW's field of action moves into the opponent's brain, or more precisely into the subconscious of the opponent's population. The aim of this war is to alter reality by, among other things, often inundating the enemy's population with false information, sometimes backed up by so-called experts, rumours and fabricated or modified videos. What's more, the proliferation of social bots means that disinformation can be automatically generated on social networks. For example, 400,000 bots are said to have automatically generated around 19% of the total volume of tweets during the 2016 US presidential campaign.
In TRADEF, which is part of the ASTRID call for projects targeting cognitive warfare, we are proposing to tackle a number of areas of disinformation: fake news and deepfake. The idea is to detect very quickly in social networks, the birth of a fake in its textual, audio or video form and its propagation through the networks. Unlike Botsentinel, which uses Twitter accounts to classify them as trustworthy or not by storing and tracking them daily, TRADEF takes a completely different approach. The aim is to detect the birth of a 'fake' and track it over time. At any given moment, this potential rumour is analysed and assigned a confidence rating, by tracking it across social networks in the reference language as well as in social networks where the language is different from the one chosen. The evolution of the suspect information over time will see its score change according to the data with which it is confronted. This data can be matched with audio or video data that can confirm or refute the credibility of the information to be processed. Videos that can be used as sources for a fake can themselves be deepfakes. This leads us to be
vigilant in examining these videos by developing robust deepfake detection methods. Indeed, according to the various international campaigns evaluating these methods, it is possible to obtain high identification rates on the baselines used, but the results deteriorate drastically on new data, as will be shown later in this project. Finally, a dimension of explicability of the results is introduced in this project, making it possible to explain the process that led to the affirmation or non-assertion of the status of the event at a given moment. Given the experience of the participating teams in deep learning and automatic processing of the standard Arabic language and its dialects, we propose to track down and identify fakes and potentially harmful information in Arabic social networks, which will give rise to other interesting scientific challenges, such as the processing of code-switching,
the variability of Arabic dialects, the identification of named entities in the speech continuum, the development of neural methods for languages with few resources and the explicability of the results obtained.
Reference
ANR-22-ASGC-003-02
Website
Programme
Europe Creative
Sub-programme
CREA-CULT-2023-COOP-1
Duration
3 years
Scientific manager
Paola Ranzini
Laboratory
Cultural Identity, Texts and Theatricality Laboratory
Project coordination
Avignon University
Partners
National and Kapodistrian University of Athens (EL) - Associazione Culturale PACTA Arsenale dei Teatri (IT) - Multicultural city e.V. (DE) - Libera Università di Lingue e communicazion IULM (IT) - Theatro Atomon Me Anapiria (EL)
Official summary
TraNET was born out of an observation of the forms of theatrical consumption in the current European theatrical landscape. The project aims to
implement new strategies for the internationalisation of live theatre using audiovisual tools to increase access to and participation in transnational culture.
Transnational culture, and audience involvement and development - both physically and digitally. The challenge of the project is to
tackle the fragmentation of national and linguistic theatrical consumption by enabling new modes of transnational circulation of theatrical performances (international audiences) and creative professionals.
(international audience) and creative professionals (creative exchanges). In order to be active at an international level, the project seeks to include grassroots organisations.
The project seeks to include grassroots organisations and micro-organisations in capacity building within Europe's cultural and creative sectors.
European cultural and creative sectors.
The project focuses on the biennial edition of a transnational theatre festival during which three new plays, revivals of national classics on the EU's transversal priorities, and a new play will be presented.
and a new multilingual European co-production. The plays will be performed in person in one country, with simultaneous screenings streamed from abroad.
in one country and simultaneous streaming of live theatre performances in other countries. The new digital performance strategy
live multilingual surtitling and interpreting will enable spectators to become part of a virtual transnational audience, involved in a multilingual show and in a new multilingual European co-production.
a virtual transnational audience taking part in a multilingual performance and a post-show debate managed by the production companies involved.
The national classics will be selected on the basis of the following themes: (i) democracy to empower students from remote/disadvantaged regions; (ii) equality for
(ii) equality to inspire minority groups who have not yet been included; (iii) parity to prevent gender discrimination.
The transnational co-production will be a differentiated and multilateral theatrical adaptation of the poem The Waste Land by T.S. Eliot.
Each of the theatre companies involved will project their own perspective and point of view as they interpret this classic European masterpiece.
Reference
101132043
Programme
ANR - Specific
Sub-programme
Sciences of matter and engineering
Type of project
PRC
Duration
4 years
Scientific manager
Christine PEPIN
Laboratory(ies)
UPRI
Project coordination
CNRS Hauts-de-France
Partners
Sorbonne University - National Centre for Scientific Research
Summary
The project aims to investigate the use of sonication on micrometric drug vectors to significantly increase payload and drug release rates compared to current strategies.
Since active substances can be hydrophobic or hydrophilic, we will consider both vectors consisting of oil-filled droplets stabilised by surfactants, and vectors filled with water and stabilised by a cross-linked or gelled membrane. In all
In this case, the sonication must be adjusted to avoid cavitation inside the microdroplets (MD).
The aim is to study how ultrasonic stimulation of the drug carrier (DC) influences fluid and drug movement, and how drugs can best be used in microdroplets to reach exovascular tissue through the permeable vessel wall.
The aim of the project is to provide advanced experimental and numerical modelling techniques to understand sonication release from the wall of a permeable vessel.
Indeed, many quantities/phenomena cannot be measured/studied experimentally (e.g. internal membrane stresses, damage formation) and need to be studied as part of a research project. These results will complement the major experimental campaigns of the project and will help to provide a fundamental understanding of the coupling between (i) the DC deformation under ultrasonic stimulation and the hydrodynamics induced by the confined flow of the external fluid,
and (ii) mass transfer of the drug out of the DC and into the surrounding tissues.
Programme
ANR- Cross-disciplinary areas
Sub-programme
Interfaces: digital sciences - humanities and social sciences
Type of project
Collaborative research project
Duration
4 years
Scientific Manager
Tania JIMENEZ
Laboratory
UPR 4128 LIA - Avignon Computing Laboratory
Project coordination
Partners
CRAN (Centre de recherche en automatique de Nancy) - Avignon University - ESCE (Ecole supérieure du commerce extérieur)- FPF(Financement participatif France)
Summary
The UMICrowd project explores crowdfunding from an economic and sociological perspective, using mathematical modelling, artificial intelligence (AI) and empirical analysis tools. It aims to provide decision-support tools to help entrepreneurs design their campaigns and crowdfunding platform managers select, rank and promote projects.
Reference
ANR-22-CE38-0013-03
Website
Programme
LIFE
Sub-programme
Climate Action
Duration
4 years
Responsible scientifique
Armin BISCHOFF
Laboratory(ies)
UMR 7263 IMBE Mediterranean Institute of Biodiversity and Ecology EECAR
Project coordination
Landgesellschaft Sachsen (DE)
Partners
HBLFA Raumberg-Gumpenstein (AT) - Ökológiai Mezőgazdasági Kutatóintézet (HU) - Hochschule Anhalt (DE) - Steirisches Landesweingut Silberber (AT) - Marrenon Vignobles en Luberon & Ventoux (FR) - Landesweingut Kloster Pforta ( DE)
Summary
Winegrowing in Europe is having to cope with the effects of climate change. The increase in summer droughts, the high risk of erosion due to heavy rainfall and the arrival of new parasites require innovative solutions.
The aim of the LIFE VineAdapt project is to help improve the resilience of wine-growing ecosystems in the face of climate change. Richer biodiversity and better adapted vineyard management are crucial. This project builds on the results of the LIFE VinEcos project that preceded it.
The LIFE VineAdapt project will run for 5 years. Eight scientific and technical partners from Germany, France, Austria and Hungary will focus on the project's five themes until June 2025. The transfer of knowledge between researchers, vineyard technicians and winegrowers is a key aspect of this project.
Reference
LIFE19 CCA/DE/001224
Website
Programme
ANR - Specific
Sub-programme
ANR JST CREST
Type of project
International collaborative research project
Duration
5 years
Scientific manager
Jean-François BONASTRE
Laboratory(ies)
UPR 4128 LIA Laboratoire Informatique d'Avignon
Project coordination
Avignon University
Partners
EURECOM - National Institute of Informatics
Summary
Thanks to recent advances in automatic speech and language processing, humans are increasingly able to interact by voice with intelligent artificial agents. The number of applications using voice in this way is growing rapidly, and this mode of interaction is becoming increasingly accepted. Today, voice systems can provide synthesised messages of such quality that they are difficult to distinguish from messages recorded by a human. They are also capable of understanding requests expressed in natural language, while remaining within their precise application framework. Finally, these systems frequently recognise or identify their users by their voice.
This project focuses on the notion of voice identity. This primarily concerns voice generation and speaker recognition. Voice generation refers to the set of speech interface modules that can be used to produce speech extracts that sound like a given natural voice. These modules include speech synthesis and voice conversion technologies, both of which can produce voice samples corresponding to the vocal identity of a targeted person. Speaker recognition is part of voice biometrics and consists of determining or verifying a person's identity through their voice. Voice generation and speaker recognition are two technologies that can generate conflicts, either with each other or with other aspects of voice interfaces. Voice generation seeks to artificially produce speech that sounds "natural" and produced by a given person, whereas speaker recognition seeks to verify the authenticity of a voice message and the identity of the person who produced it. A speaker recognition system may be used to train a voice generation system, with the result that the synthetic voice produced by the final system may mislead the voice biometrics system... A second conflict arises when speaker recognition is used without the speaker's knowledge. To protect against this, a "voice anonymisation" approach, which involves both speaker recognition and voice generation aspects, needs to be developed to remove a speaker's identity from a voice message while preserving at least its linguistic content but also its natural aspects, its emotionality, its "colour", etc. The three aspects - generation, identity recognition and anonymisation - are closely linked and need to be considered together.
VoicePersonae aims to bridge the technological gap between the different aspects of the notion of "voice identity" presented above. This project proposes to (a) model "voice identity" (b) improve the security and robustness of voice biometric systems (c) protect the privacy of users. VoicePersonae will bring together disparate approaches to multi-speaker voice generation, combining voice synthesis and transformation. To achieve this, VoicePersonae will exploit the latest speaker recognition technologies. VoicePersonae will enhance the security and robustness of voice biometrics by exploiting the results obtained to counter attacks using voice generation. This will be achieved by assuming that fraudsters are familiar with the technologies we employ, using the finesse of the "voice identity" modelling developed in this project. Finally, VoicePersonae will offer the first explicit voice anonymisation solution to protect personal data. To stimulate this field of "voice identity" and more specifically the task of voice anonymisation, VoicePersonae will organise the first open challenge on the anonymisation and re-identification of speech.
Reference
ANR-18-JSTS-0001-02
Website
Programme
ANR- Generic
Type of project
PRC
Duration
5 years
Scientific manager
Jean-François BONASTRE
Laboratory(ies)
UPR 4128 LIA Laboratoire Informatique d'Avignon
Project coordination
Avignon University
Partners
Aix-Marseille Univerité - Phonetics and Phonology Laboratory - National Metrology and Testing Laboratory - Central Technical and Scientific Police Department
Summary
VoxCrim concerns the identification of individuals by their voice in the forensic field and, more precisely, "voice comparison" for national security and forensic expertise. It is fully in line with Challenge 9, and more specifically with Axis 3. VoxCrim proposes an objective and validated scientific framework for voice comparison, regardless of the type of method used (automatic or phonetic). It proposes to scientifically objectify the possibilities of implementing a voice comparison. This project brings together specialists in computer science (LIA), phonetics (LPL, LPP) and standardisation (LNE) and representatives of law enforcement agencies (SDPTS). The IRCGN will also be asked to act as a centre of expertise. VoxCrim's objectives are based on two pillars: (1) rapid exploitation of the results for well-controlled situations and (2) extending the method's operability over the longer term. (1) The first pillar is based in particular on the results of the ANR Fabiole project. It uses the methodology and database and extends the expectations to define the innovative "box-rule" concept, i.e. a perfectly defined set of conditions for implementing voice comparison within which certification is possible. The technological tools and database developed within Fabiole will be improved, extended and packaged so that they can be disseminated to consortium partners. (2) The second pillar aims to add dimensions to the conditions for implementing voice comparison. This involves taking into account voice characteristics that are closely linked to the 'field' context of police services, such as the influence of the socio-cultural and linguistic environment (mother tongue, family language, etc.). To this end, a database (PTSVox) recorded by the SDPTS in a microphone condition and a GSM condition has been created and will be used in this project. The strategy adopted consists of developing two complementary types of analysis: a) an acoustic analysis of the productions aimed at identifying relevant clues for characterising the voices, and b) perceptual experiments aimed at assessing the listeners' ability to discriminate between voices. The acoustic analyses will delimit inter- and intra-speaker variability in order to test the robustness of these markers. The perceptual experiments will show whether the acoustic descriptors highlighted above are actually used by listeners to identify speakers, particularly for telephone speech (the most common case in forensic cases), and will assess a listener's performance in terms of voice discrimination. Ultimately, this project will enable forensic science laboratories to add voice comparison to their 17025-type forensic science laboratory standardisation approach. Voxcrim's ambition is to disseminate the principles and interests of a quality approach in the field of voice comparison to the speech community as well as to players in the legal system (police, magistrates). To this end, theme days will be organised to disseminate the knowledge and issues addressed to the public concerned. In addition, doctoral and post-doctoral students recruited from partner laboratories will undertake 'practical' placements at the SDPTS and the IRCGN to gain an understanding of the issues faced by these operational services. In this way, VoxCrim will train a large group of scientists specialising in voice comparison in the forensic field, thereby making up for a weakness often noted in France.
Reference
ANR-17-CE39-0016-01
Website
Programme
ANR - Generic
Type of project
PRC
Duration
3 years and 7 months
Scientific manager
Laure CASANOVA
Laboratory(ies)
UMR 7300 ESPACE Study of Structures, Adaptation Processes and Changes in Space
Project coordination
CNRS IDF WEST&NORTH
Partners
Centre Max Weber - Géographie-cités - Studies of structures, adaptation processes and changes in space
Summary
WIsDHoM analyses the growth in socio-spatial inequalities caused by house price inflation. In most cities, house prices have risen faster than incomes. Property has thus become a key component of inequality, with households increasingly concentrating their investments and wealth in this sector. A CRP (CES 41, challenge B8, axis 2) has been asked to study income and wealth inequalities using a multidisciplinary approach (geography, spatial planning, history, political science, sociology, economics) and to understand the role of property in reinforcing spatial and wealth inequalities.
WIsDHoM brings together housing experts to analyse the systemic nature of inequalities linked to the dynamics and the political and financial context of the French market since the late 1990s. Over the last twenty years or so, French conurbations have been characterised by a sustained rise in prices coupled with a steady increase in the proportion of homeowners. This situation, described as a 'robust bubble', lends credence to the idea of a new regime of high prices likely to explain the resilience of the markets, which is paradoxical given the deterioration in household purchasing power and in the profitability of rental properties following the 2008 crisis. This regime combines factors at different levels: national (conditions of access to credit, incentives for home ownership, the growing role of property ownership in household wealth and as a form of protection replacing social protection systems) and local (policies to stimulate the market and spatial differentiation of property prices).
The aim of the WISDHOM project is to understand the interactions between local price dynamics, inequalities and policies. The variability of house prices, which depends on a number of local conditions, determines the social sorting of property owners, the conditions for (de)valuing household assets, and the property opportunities for local players, both public and private. In turn, these factors influence market conditions. It is the spatial study of this self-reinforcing effect that forms the core of the project.
The study is based on three cases at three levels of the urban hierarchy: Paris, Lyon and Avignon. The empirical analysis consists of a multiscalar approach to understand the links between national and local policies and household strategies. Our scientific programme has three main strands: 1) an in-depth analysis of the social construction of local housing markets by public and private players in the three selected conurbations 2) a spatial analysis of inequalities in access to housing using disaggregated databases (property transactions, taxation, mortgages, land registry);3) A household survey of property ownership strategies in our three areas, with the aim of highlighting the forms of enrichment and vulnerability associated with property ownership on the one hand, and the way in which these strategies (housing renovation, bimby, nimby, etc.) can have repercussions on housing policies on the other.) can have an impact on local housing policies. The team's multi-disciplinary expertise in survey and data processing will enable it to meet the challenge of integrating sources that have never before been treated jointly.
Reference
ANR-18-CE41-0004-02
Website
Contact
Projects, Partnerships and International Development Division
aap-recherche@univ-avignon.fr
Mis à jour le 2 April 2024