Archives

Position on data-driven weather forecasting

Offre en lien avec l’Action/le Réseau : – — –/– — –

Laboratoire/Entreprise : Centre National de Recherches Météorologiques
Durée : 21 months
Contact : laure.raynaud@meteo.fr
Date limite de publication : 2024-01-31

Contexte :
In the past months several studies have proposed a new paradigm for weather prediction, demonstrating that data-driven forecasting models are now competitive with traditional physics-based models, on a number of variables and at a relatively coarse scale.
In this context, the main mission of the open position is to carry out innovative work to develop an emulator of numerical weather prediction model on a regional scale, based on state-of-the-art artificial intelligence methods (transformers, graph networks, generative models).
Application on CNRS portal : https://emploi.cnrs.fr/Offres/CDD/UMR3589-LAURAY-005/Default.aspx

Sujet :
In the current state of knowledge and progress, the areas to explore are:
1. the creation of learning datasets, based on simulations and/or observations;
2. the choice of a neural network architecture adapted to the problem, based on recent literature on the subject (transformers, graph networks, etc.);
3. implementation and training of the selected architecture on the available datasets;
4. evaluation of the emulator, using diagnostics and objective forecast scores, and analysis of high-impact weather situations.
The developments will be carried out in priority for high-resolution forecasting on a regional scale, over France and neighboring countries, but could be extended to forecasting on overseas areas or for forecasting on a global scale. The work will use the most innovative aspects of deep learning, and will require scientific watch on these techniques. The position holder will also be required to collaborate with other Météo-France services, scientists and industrials in the mathematical and IT fields, and European partners. He or she will be encouraged to participate in international institutional or scientific meetings, to contribute to national or European projects, and to participate to the publication of results in the form of scientific articles.

Profil du candidat :
We are seeking for a highly motivated candidate with a strong experience in deep learning with large-scale data.

Formation et compétences requises :
The ideal candidate will have the following qualifications:
– A master’s or engineering degree in computer science or statistics
– Recognized expertise in statistical learning methods (convolutional neural networks, Transformers, graph networks, GANs, etc.)
– Excellent capabilities in Python programming and classic development environments for deep learning (PyTorch, Tensorflow)
– Experience in using GPU processors and code optimization (parallelization)
– Experience in handling very large datasets
– An ability to work scientifically in a team, with internal and external collaborators, and to communicate results in English, both written and orally
– Scientific curiosity, rigor, autonomy
– An interest in the application to atmospheric sciences; a first experience in this field will be an advantage.

Adresse d’emploi :
CNRM, Toulouse, France.
The National Center for Meteorological Research (CNRM) is a Joint Research Unit (UMR) with dual supervision from Météo-France and CNRS. The CNRM conducts research in conjunction with the missions of Météo-France and other organizations, notably the CNRS and universities. The CNRM participates in major international or European scientific programs in its field of expertise.
The position holder will be part of the Predictability team, located on the Météopole site in Toulouse. The team’s research focuses on the development of operational ensemble forecasting systems at Météo-France and the integration of AI methods in atmospheric modeling.

Detection of electric arcs using physics-informed neural networks

Offre en lien avec l’Action/le Réseau : BigData4Astro/– — –

Laboratoire/Entreprise : DAVID Lab – UVSQ – Versailles
Durée : 36
Contact : mustapha.lebbah@uvsq.fr
Date limite de publication : 2024-01-31

Contexte :
An electric arc is a high-current disruptive discharge capable of self-sustaining at low voltage. Its study belongs to the field of plasma physics.

While legacy networks have learned to live with the non-zero probability of this fault occurring, since the causes and consequences are known, the same cannot be said for networks dedicated to propulsion, whether all-electric or hybrid. The criticality of the damage is often equated with the energy deployed by the fault compared with the susceptibility to damage of the materials in direct or indirect contact with the arc. We can therefore imagine the impact that a high-power arc could have in such a network (KV/MW) if its duration were to exceed a few milliseconds! Given the increase in voltage and the DC waveform, the risk and consequences of arc faults are increased. This observation, combined with a constrained environment (confined areas, high-risk zones such as FFLZ, severe environment zones, etc.), prompted the EWIS engineers to carry out high-power electric arc tests, leading to the following conclusions:

When an electric arc is generated whose power exceeds several hundred kW and whose lifetime is not controllable, it is no longer possible to control it.

When an electric arc is generated, the power of which exceeds several hundred kW and the lifetime of which cannot be controlled, it is no longer possible to mitigate the consequences solely by choosing ‘arc-resistant’ materials and design guides in the confined environment and safety constraints inherent in civil aeronautics.

This is why the detection and elimination of this fault is inevitable. Arc detection systems, which were not mandatory on legacy networks, will certainly be required on the propulsion networks envisaged.

Arc detection is important for electrical safety, because arcs can cause fires, damage electrical equipment and pose a risk to people. To detect arcs, several technologies and methods are used, including: monitoring by current sensors, light sensors, electrical signal analysis, thermal cameras, etc.

Sujet :
Objectives
The aim of the proposed thesis will be to couple AI and physics for the modeling and early detection of electric arcs. The AI tools targeted are Physics-Informed Neural Networks (PINNs). Experimental data will be available to feed these models. In addition, the physical equations to describe the evolution of arcs to be taken into account are well-known (Maxwell, Faraday, Navier-Stokes, etc).

The major challenges of the thesis will be as follows:

• Modeling electric arcs: Developing PINNs models to describe electric arcs accurately, taking into account the physical equations that govern them. As a check on the calculations (or as a learning tool), physical simulations called MHD (Magneto-Hydro-Dynamics) could be carried out for all the geometries envisaged in the SafranTech E&E team.
• Network training: Train neural networks to predict the presence of electric arcs using observation data and the relevant physical equations.
• Early detection: Develop techniques for the early detection of electric arcs based on PINNs models.
• Experimental validation: Test PINN models and detection methods on real experimental data from Safran electrical systems. This data will be available for several “typical” geometries and will enable the PINNs models to be tested on reproducible and controllable cases.

Profil du candidat :
● End of engineering degree / Master’s degree in a relevant field (e.g., computer science, ML/AI, Statistics …)
● Excellent understanding of machine learning and physics basics. Familiar with recent Artificial Intelligence: transformers, diffusion
model, auto-encoder…etc.
● Excellent programming skills, especially with Python, Pytorch,
● Autonomous and able to quickly adapt to recent scientific literature / technologies.

This PhD topic is participating to the Université Paris-Saclay EU COFUND DeMythif.AI program : https://www.dataia.eu/actualites/cofund-demythifai-appel-sujets-de-these. It is reserved to international students who have spent less than 12 months in France in the last 3 years. The candidates will be evaluated by a jury that will select 15 PhD to start in fall 2024. The successful candidates will be fully funded for 3 years, have access to specific scientific and non-scientific training, and be fully part of the Université Paris-Saclay AI community. The aim of this Ph.D. research is to strengthen collaboration with Safran group. The thesis is accompanied by a collaborative contract with Safran, ensuring the environment, interaction with experts, and data availability.

Formation et compétences requises :
● End of engineering degree / Master’s degree in a relevant field (e.g., computer science, ML/AI, Statistics …)
● Excellent understanding of machine learning and physics basics. Familiar with recent Artificial Intelligence: transformers, diffusion
model, auto-encoder…etc.
● Excellent programming skills, especially with Python, Pytorch,
● Autonomous and able to quickly adapt to recent scientific literature / technologies.

This PhD topic is participating to the Université Paris-Saclay EU COFUND DeMythif.AI program : https://www.dataia.eu/actualites/cofund-demythifai-appel-sujets-de-these. It is reserved to international students who have spent less than 12 months in France in the last 3 years. The candidates will be evaluated by a jury that will select 15 PhD to start in fall 2024. The successful candidates will be fully funded for 3 years, have access to specific scientific and non-scientific training, and be fully part of the Université Paris-Saclay AI community. The aim of this Ph.D. research is to strengthen collaboration with Safran group. The thesis is accompanied by a collaborative contract with Safran, ensuring the environment, interaction with experts, and data availability.

How to apply
https://adum.fr/as/ed/voirproposition.pl?site=PSaclay&matricule_prop=51640&langue=en

Adresse d’emploi :
David Lab /UVSQ Versailles

Document attaché : 202312150810_Proposal-PHD.pdf

optimization in the presence of uncertainties, application to the energy efficiency of buildings

Offre en lien avec l’Action/le Réseau : RoCED/– — –

Laboratoire/Entreprise : Ecole Centrale Lyon
Durée : 12 months
Contact : celine.helbert@ec-lyon.fr
Date limite de publication : 2024-03-31

Contexte :
Including model and environmental uncertainties in decision aiding methods is often seen as becoming increasingly important. This is the case when seeking optimal renewing strategies for buildings.
However the theory and the algorithms for optimizing in the presence of uncertainties is still an active research domain, particularly when optimizing many criteria.
In this post-doctoral work, we will focus on costly and general nonlinear constrained multi-objective optimization problems that are affected by uncertainties. We will consider the case where the uncertain parameters can be separated from the optimization variables and can be chosen during the simulations. Because of this separation and providing a probability of occurence of the uncertainties exists, a statistical modeling in the joint design × uncertain parameters space is possible. This will be the context of the work.

Sujet :
The goal of this work is to improve the ideas introduced in [El Amri 23] by putting them in the context of multi-objective optimization under uncertainties. The expected hyper-volume improvement must be adapted to take into account the uncertainties and a sampling SUR criterion must be devised to choose the value of the random parameter to be evaluated. A multi-output Gaussian process can be proposed to take into account the correlation between the objective functions. A wise choice of the correlation kernel should be done.
The methods developed will be applied to the design of energy efficient
buildings, a major contemporary challenge. The criteria are the energy usage of the building, the thermal comfort and the cost. Important uncertainties affect the cost (through the cost of energy) and the external conditions through the climate change.
[El Amri] : R. El Amri, R. Le Riche, C. Helbert, C. Blanchet-Scalliet and S. Da Veiga, A sampling criterion for constrained Bayesian optimization with uncertainties, to appear in SMAI Journal of Computational Mathematics, 2023.

Profil du candidat :
• doctoral degree or equivalent in mathematics,
• proven strong background in uncertainty quantification or statistical learning theory,
• substantial experience in numerical programming.

Formation et compétences requises :
See above

Adresse d’emploi :
Institut Camille Jordan (ICJ), Campus of l’Ecole Centrale de Lyon, Ecully.
Stays will be expected at the LIMOS laboratory, either in Clermont-Ferrand, or in Saint-Etienne, FR.

Document attaché : 202312141710_postdocoffer_moo_uncertainties.pdf

Synergies in Turbulent Natural Convection: Bridging Convolutional Neural Networks, Physics- Informed Machine Learning, and High-Performance Computing for improved modeling

Offre en lien avec l’Action/le Réseau : – — –/– — –

Laboratoire/Entreprise : LISN – UMR9015
Durée : 12 (+6)
Contact : anne.sergent@lisn.fr
Date limite de publication : 2024-06-30

Contexte :
The mechanical engineering department of the LISN lab invites applications for a one-year postdoctorate position to conduct cutting-edge research at the intersection of turbulent natural convection, convolutional neural networks (CNN), physics-informed machine learning, and high-performance computing (HPC). The successful candidate will work on advancing the field of super-resolution analysis for turbulent fluid flows using innovative approaches based on numerical and
experimental ombroscopy techniques.

Supervision and research team

The Postdoc will work in collaboration with Didier Lucor and Anne Sergent from LISN, and Julien Salort and Francesca Chillà from the Physics Lab of ENS Lyon (https://www.ens-lyon.fr/PHYSIQUE). Thus, the research team is composed by physicist, fluid mechanics and artificial intelligence researchers from different laboratories, leading to a multidisciplinary project funded by ANR.

Funding

This project is funded by the ANR research project THERMAL.
The post-doctoral position is a one-year full-time appointment starting during 2024. Gross salary will depend on the experience of the candidate, up to approx. 40,000 €/year (net salary: up to approx. 32,000 €/year). The candidate will also benefit from French social insurance.
Within the framework of the ANR project THERMAL the postdoc will have funding for participation in conferences, publication fees and visits to Lyon lab. Moreover, the postdoc will have access to compute servers from University Paris-Saclay and GENCI national supercomputers.

Deadline for Applications: first semester 2024
The Postdoc is expected to start in 2024 (preferably during the first semester)

Application Process
Interested candidates should submit the following documents to didier.lucor@lisn.fr and anne.sergent@lisn.fr :
1. Curriculum Vitae (CV) including a list of publications.
2. Cover letter detailing the candidate’s research experience and interest in the position.
3. Contact information for three references.

Sujet :
The research will build upon recent surveys on machine-learning-based super-resolution reconstruction of turbulent flows. The candidate will explore and develop methods to enhance the resolution of turbulent flows through the application of CNN-based techniques, physics-informed loss
functions with access to direct numerical simulations databases produced with high-performance computing technologies on national supercomputers. The goal is to reconstruct instantaneous vortical
flows and temperature fields with high fidelity, even in scenarios with limited/partial training data and noisy inputs.

Key Responsibilities

1. Implement and refine machine-learning models, particularly CNN-based methods, for super-resolution reconstruction of turbulent flows.
2. Investigate the use of physics-informed loss functions and neural network structures to improve the accuracy and robustness of super-resolution models.
3. Collaborate with the lab team to integrate multi-scale filters, unsupervised techniques, and spectral properties into the super-resolution models.
4. Assess the robustness and sensitivity of models against noisy inputs, especially in the context of experimental measurements.
5. Contribute to the development of super-resolution models in wavespace for incorporating specific spectral properties.

Profil du candidat :
– Ph.D. in Computational Fluid Mechanics, Aerospace Engineering, Applied mathematics, Computer Science or a related field.
– Proven track record of publications in relevant peer-reviewed journals.

Formation et compétences requises :
– Strong background in machine learning, particularly convolutional neural networks.
– Experience in physics-informed machine learning and high-performance computing.
– Very good programming skills (e.g., Python, TensorFlow, PyTorch).

Adresse d’emploi :

Page d’accueil

LISN lab (CNRS & Université Paris Saclay):
The mechanical engineering department develops broad-spectrum research activities mainly in fluid mechanics and computer science. Over the last decade, expertise has developed at the interface of computational fluid mechanics, HPC and physics-informed machine learning, uncertainty
quantification and data assimilation techniques.

Document attaché : 202312141407_postdoc-anr-thermal_v2.pdf

Experimenting Embeddings with Graph Neural Networks for Knowledge Graphs using RDF Reification

Offre en lien avec l’Action/le Réseau : – — –/– — –

Laboratoire/Entreprise : Université de Nantes
Durée : 5 à 6 mois
Contact : Patricia.Serrano-Alvarado@univ-nantes.fr
Date limite de publication : 2024-03-31

Contexte :
The context of this work is the CLARA project [CLARABench]1. In this internship, we are interested in analyzing knowledge graphs using deep learning methods. Since their introduction, deep learning models have been at the center of attention. The latest examples are the Large Language Models (LLMs) and in particular the transformer model [Transformer] used by ChatGPT. These models are built for tasks such as chatbots, conversational AIs, or sentiment analysis of texts. However, deep learning models have proven to be very efficient for other tasks, like the Convolutional Neural Networks (CNNs) for image recognition. This efficiency has also been proven for analyzing knowledge graphs with Graph Neural Networks (GNN) [SurveyGNN]. GNN models are particularly efficient for tasks such as link prediction, entity classification, or k-nearest neighbours.
Knowledge graphs represent statements as triples (head, relation, tail). Each triple is a fact stating a relation between two entities. Statements about statements, also called statement-level annotations, are increasingly used. They allow specifying that a statement is true under a particular context. Context can concern temporal aspects, provenance, trust values, scores, weights, etc. RDF reification allows expressing statement-based annotations in a generic way. We believe that current GNN models are not suited for processing annotations. GNN models do not support RDF reification because it may introduce noise that would reduce the quality of the results.

Sujet :
The goal of this internship is to show the limits of existing GNN models in the presence of RDF reification and to propose a new model that efficiently integrates RDF annotations.

You will participate in research work alongside a PhD student in the following tasks:
● Defining and running an experimental protocol. You will conceptualize and run an experimental protocol to put into light the impact of reification on GNN models. This will require the understanding of (a) several models (e.g., R-GCN [RGCN], HypE [HypE], RDF2vec [RDF2vec]), (b) the different reification approaches and their impact on GNN models (standard reification [Standard], n-ary relations [N-ary],

RDF-star [RDF-star]), and (c) how the impact of RDF reification on the GNN models can be measured. The obtained results will be the baseline for the next task.
● Creating a new GNN model. You will help in contributing a GNN model that better integrates RDF reification. The model will be inspired by the message-passing algorithm used in GNNs such as R-GCN and it should be able to adapt to RDF reification. The result of this approach should be compared to the baseline previously obtained.

Profil du candidat :

Knowledge of Machine Learning and Deep Learning.

Good programming skills, in particular in Python.

Formation et compétences requises :
Master or Engineer Student.

Adresse d’emploi :
Université de Nantes

Document attaché : 202312141033_2023-2024 Stage Master 2.pdf

Méthodes de dé-mélange pour la correction d’atténuation en tomographie optique diffuse de fluorescence

Offre en lien avec l’Action/le Réseau : – — –/– — –

Laboratoire/Entreprise : Insitut Fresnel
Durée : 5 mois (Mars à Juill
Contact : andre@fresnel.fr
Date limite de publication : 2024-03-31

Contexte :

Les technologies d’imagerie capables de détecter les processus biologiques précoces in vivo de manière non invasive pour des études longitudinales, avec une haute résolution, représentent un défi pour la recherche biomédicale. Le concept de notre système d’imagerie repose sur un nouveau d’imagerie optique diffuse de fluorescence multicolore pour l’imagerie in vivo du petit animal en trois dimensions (3D) dans la fenêtre NIR-II (1000-2000 nm). La tomographie optique diffuse de fluorescence consiste à injecter au sujet (ici une souris) des substances chimiques qui se fixent sur différents organes. Ces substances chimiques, appelées fluorophores, sont alors excitées par une source lumineuse puis réémettent de la lumière lors de leur relaxation, à plus faible énergie (plus longue longueur d’onde). L’objectif est de reconstruire des images à partir de ce signal de fluorescence. Le signal de fluorescence ainsi que la source d’excitation peuvent être atténués à la fois par l’absorption et la diffusion des différents milieux traversés, ce qui entraîne une distorsion des spectres mesurés. Les méthodes conventionnelles de dé-mélange linéaire permettent de séparer les spectres sans tenir compte de ces effets.

Sujet :
Les algorithmes de dé-mélange multilinéaire [1] ont montré leur efficacité pour la séparation de signaux multidimensionnels issus de la spectroscopie de fluorescence [2]. Ils permettent d’estimer les spectres d’excitation, les spectres d’émission de fluorescence ainsi que les concentrations relatives de plusieurs fluorophores présents dans plusieurs solutions chimiques.
Le but de ce stage est de mettre en œuvre des méthodes de décomposition multilinéaire pour corriger des images tomographiques hyper-spectrales de l’atténuation [3].

[1] T. G. Kolda and B. W. Bader, “Tensor decompositions and applications,” SIAM Review, vol. 51, no. 3, pp. 455–500, 2009.

[2] R. Bro, “Parafac, tutorial and applications,” Chemom. Intel. Lab. Syst., vol. 38, pp. 149–171, 1997.

[3] Hayato Ikoma, Barmak Heshmat, Gordon Wetzstein, and Ramesh Raskar, “Attenuation-corrected fluorescence spectra unmixing for spectroscopy and microscopy,” Opt. Express 22, 19469-19483 (2014)

Profil du candidat :
Le candidat devra être particulièrement à l’aise en programmation (python/Matlab) et avoir une réelle appétence pour les interactions entre l’informatique et la physique.

Formation et compétences requises :
Le candidat recruté devra être en dernière année d’école d’ingénieurs ou en Master 2 dans le domaine des mathématiques appliquées, le traitement du signal/images ou dans une formation équivalente. Il devra être particulièrement à l’aise en programmation (python/Matlab) et avoir une réelle appétence pour les interactions entre l’informatique et la physique.

Adresse d’emploi :
52 Av. Escadrille Normandie Niemen, 13013 Marseille

Automatic classification of plasmodium parasite species and stages of development from stained thin blood smears using machine learning

Offre en lien avec l’Action/le Réseau : – — –/– — –

Laboratoire/Entreprise : Centre d’épidémiologie et de santé publique des a
Durée : 4-6 mois
Contact : muriel.visani@univ-lr.fr
Date limite de publication : 2024-03-31

Contexte :
Voir le fichier ci-joint

Sujet :
Voir le fichier ci-joint

Profil du candidat :
Voir le fichier ci-joint

Formation et compétences requises :
Voir le fichier ci-joint

Adresse d’emploi :
Centre d’épidémiologie et de santé publique des armées (CESPA), Marseille.

Document attaché : 202312131521_InternshipBloodSmear-CESPA-final_compressed.pdf

CORIA 2024 (COnférence en Recherche d’Information et Applications)

Date : 2024-04-03
Lieu : La Rochelle

CORIA (COnférence en Recherche d’Information et Applications) est la principale manifestation soutenue par l’Association Francophone de Recherche d’Information et Applications ARIA (http://www.asso-aria.org ).

Dates importantes

Soumission des articles résumés, courts et longs : jeudi 1er février 2024

Notification aux auteurs : mardi 5 mars 2024

Conférence : les 3 et 4 avril 2024 à La Rochelle

CORIA vise à rassembler les équipes et les personnes menant des travaux scientifiques dans le domaine de la recherche d’information et de ses applications : recherche d’information sur le web, sur les réseaux sociaux ou sur des collections spécifiques, systèmes de recommandation, fouille de documents, d’images, d’enregistrements audio, de vidéos, assistants personnels et chatbots… Devenue activité quotidienne du grand public, la recherche d’information est essentielle à de nombreux usages du numérique. L’activité scientifique et technologique associée ne cesse de croître en interaction avec d’autres domaines de l’informatique et d’autres disciplines, mathématiques, linguistique, sciences cognitives, mais aussi en lien direct avec l’industrie et les acteurs de l’internet, des médias, de la culture, de la santé ou de l’éducation. Les modèles récents intègrent l’apprentissage automatique, la fouille de données, le traitement automatique des langues, le traitement de la parole et du signal, l’analyse d’images ou encore l’informatique affective.

La conférence CORIA est ouverte à l’ensemble de la communauté scientifique internationale concernée par la recherche d’information du point de vue théorique comme du point de vue des applications. Le public visé par CORIA est celui des chercheurs académiques, incluant les étudiants en master et doctorat, des industriels et de tous les spécialistes du domaine. Toutes les publications CORIA sont diffusées en accès ouvert sur le site de l’ARIA et sont indexées par DBLP.

Soumissions

Les soumissions doivent être rédigées selon le style CEURART à une colonne et être soumises sous forme de fichiers PDF via le système EasyChair.

Système de soumission : https://easychair.org/conferences/?conf=coria2024

Modèle Overleaf : https://www.overleaf.com/latex/templates/template-for-submissions-to-ceur-workshop-proceedings-ceur-ws-dot-org/wqyfdgftmcfw

Modèle de présentation téléchargeable : http://ceur-ws.org/Vol-XXX/CEURART.zip

Il est possible de soumettre des articles dans 3 formats :

– résumé (2 pages + références) : traduction résumée d’un papier déjà publié, résultat négatif, prise de position, description d’un projet;

– court (8 pages + références) : résultats préliminaires ou état de l’art;

– long (12 à 16 pages + références) : article scientifique complet.

Les soumissions, anonymisées, seront évaluées par 3 membres du comité de programme.

Dans le cas de soumissions de résumés d’articles déjà publiés, elles ne devront pas être anonymisées, et seront évaluées par un membre du comité de programme. L’article d’origine doit être indiqué ote toute ambiguité.

Des articles de longueur inférieure à la limite peuvent être soumis sans que cela soit préjudiciable. CORIA accepte les articles en anglais lorsque les auteurs ne sont pas francophones, mais privilégie les articles en français quand l’un des auteurs est francophone pour les versions finales.

Thèmes (liste non exhaustive) :

– Apprentissage et fouille pour la RI : apprentissage profond, apprentissage de représentations, apprentissage d’ordonnancement, classification;

– Représentation de l’information : indexation, entités liées, multimédia, profils, bases de connaissances;

– Compréhension de requêtes : intention de recherche, suggestion de requêtes, difficulté des requêtes, adaptation aux requêtes;

– Interaction utilisateur : interrogation flexible, modélisation de l’utilisateur, du contexte et de l’usage, accessibilité, RI conversationnelle, personnalisation, RI collaborative, RI interactive;

– Systèmes question/réponse, systèmes de dialogue, chatbots classiques et ChatGPT;

– RI et Humanités Numériques;

– Traitement automatique de la langue naturelle écrite et orale pour la recherche d’information : résumé automatique, détection d’entités nommées et de relations, analyse de sentiments et fouille d’arguments;

– RI et connaissances : web sémantique, web des données, ontologies;

– RI pour les assistants personnels et/ou vocaux;

– RI multilingue : recherche d’information multilingue, traduction automatique, RI interlangue;

– Passage à l’échelle : architectures, performance, compression;

– Analyse du Web : grands graphes, utilisation de la topologie du web, citations, analyse de liens;

– Réseaux sociaux : analyse de réseaux, d’opinions, diffusion d’information, prédiction d’activités, détection d’événements;

– Filtrage et recommandation : filtrage collaborative, démarrage à froid;

– Multimédia (image, audio, vidéos, sons, musiques) et texte : indexation, recherche, catégorisation, alignement;

– Systèmes de recherche d’information dédiés : recherche d’information génomique, géographique, médicale, recherche de brevets;

– Ressources et évaluation de la RI : évaluation, bancs d’essais, métriques, expérimentations qualitatives des systèmes;

– Transparence, équité et explicabilité des systèmes de RI…

—

Liana Ermakova et Philippe Mulhem

Co-chairs du comité de programme de CORIA 2024

Notre site web : www.madics.fr
Suivez-nous sur Tweeter : @GDR_MADICS
Pour vous désabonner de la liste, suivre ce lien.

Workshop AI for Biological Imaging

Date : 2024-01-08 => 2024-01-09
Lieu : Sorbonne University, Paris
4, place Jussieu 75005
Bâtiment Esclangon

Dear all,

We are very happy to announce the:

7th Cross-disciplinary Genomics Meeting: “AI for Biological Imaging”
which will take place at Sorbonne University, Paris, January 8-9 2024.

Keynote speakers:
Haitham Shaban – University of Geneva, Lausanne, Switzerland
Christophe Zimmer – Institut Pasteur, Paris, France
Ulrike Endesfelder – Bonn University, Bonn, Germany
G. V. Shivashankar – ETH, Zürich, Switzerland
Olivier Colliot – Sorbonne Université, Paris, France
Thomas Walter – Institut Curie, Paris, France
Isabelle Bloch – Sorbonne Université, Paris, France
Matthieu Cord – Sorbonne Université, Paris, France
Susan Cox – King’s College London, London, UK
Maxime Deforet – Sorbonne Université, Paris, France

The registration is free but mandatory.

Please visit our website to discover our program and to register: https://ai4bi.sciencesconf.org

Looking forward to meeting you there,
Judith Miné-Hattab and Nataliya Sokolovska

Lien direct

Notre site web : www.madics.fr
Suivez-nous sur Tweeter : @GDR_MADICS
Pour vous désabonner de la liste, suivre ce lien.

Self-Supervised Anomaly Detection in complex-valued SAR imaging

Offre en lien avec l’Action/le Réseau : – — –/– — –

Laboratoire/Entreprise : ONERA / SONDRA, CentraleSupelec
Durée : 36 mois
Contact : chengfang.ren@centralesupelec.fr
Date limite de publication : 2023-12-07

Contexte :
Deep anomaly detection methods leverage neural networks to automatically extract crucial data features, mapping high-dimensional data into a more manageable, lower-dimensional latent space, thereby significantly enhancing anomaly detection performance. One standard method for anomaly detection is to utilize Autoencoders (AE) for data encoding and reconstruction, detecting anomalies based on reconstruction errors [S. Sinha, 20, S. Mabu, 21]. Due to the presence of speckle noise in SAR images, [M. Muzeau, 2022] proposed to denoise SAR images using the MERLIN algorithm [E. Dalsasso, 2021b] based on the noise2noise principle [J. Lehtinen, 18, E. Dalsasso, 21a]. This pre-processing step leads to better compression in the latent space, subsequently improving the detection performance. Further extension in [M. Muzeau, 23] proposed to guide the Adversarial AE (AAE) in the training process by filtering anomalies using an RX detector [I. S. Reed, 90].
On the other hand, self-supervised learning leverages pretext tasks to extract supervised information from unsupervised data, thereby learning valuable feature representations for downstream tasks such as classification, object detection, and segmentation [M. Caron, 21]. Self-supervised anomaly detection methods acquire data representations by creating supervised pretext tasks. The key to constructing these pretext tasks is to guide the model in learning a specialized representation suitable for anomaly detection, distinct from the general representation obtained through unsupervised learning.

Sujet :
This Ph.D. aims to investigate the above-mentioned methods for SAR anomaly detection, exploiting SAR diversities: polarimetric and interferometric channels [Pottier, 09], multi-bands, and multi-looks representation [A. Mian, 19]. Particular attention is dedicated to the phase information of the complex-valued SAR images, which is crucial to assessing the spectral (range-azimuth) bandwidth and keeping the coherency in polarimetric and interferometric channels. The Ph.D. student will rely on the previously developed open-source library (https://github.com/NEGU93) developed in [Barrachina, 19] for complex-valued radar data and based on Tensorflow although recent developments of the PyTorch framework now allow for processing complex-valued tensors with differentiable computational graphs. Using this library, it is possible to address and analyze any recent Machine Learning components like Autoencoders, Transformers, etc., through challenging theoretical methodologies (SAR denoising, self-supervised learning, characterization of latent spaces, etc.).

References:

• [S. Sinha, 20] S. Sinha et al., “Variational autoencoder anomaly detection of avalanche deposits in satellite SAR imagery,” in Proc. 10th Int. Conf. Climate Inform., 2020, pp. 113–119.
• [S. Mabu, 21] S. Mabu, S. Hirata, and T. Kuremoto, “Anomaly detection
using convolutional adversarial autoencoder and one-class SVM for landslide area detection from synthetic aperture radar images,” J. Robot., Netw. Artif. Life, vol. 8, no. 2, pp. 139–144, 2021.
• [M. Muzeau, 22] M. Muzeau, C. Ren, S. Angelliaume, M. Datcu and J. –
P. Ovarlez, “Self-Supervised Learning Based Anomaly Detection in Synthetic Aperture Radar Imaging,” in IEEE Open Journal of Signal Processing, vol. 3, pp. 440-449, 2022.
• [M. Muzeau, 23] M. Muzeau, C. Ren, S. Angelliaume, M. Datcu and J. . -P.
Ovarlez, “Self-Supervised SAR Anomaly Detection Guided with RX Detec-
tor,” IGARSS 2023 – 2023 IEEE International Geoscience and Remote Sensing Symposium, Pasadena, CA, USA, 2023, pp. 1918-1921.
• [J. Lehtinen, 18] J. Lehtinen et al., “Noise2Noise: Learning image restoration without clean data,” in Proc. 35th Int. Conf. Mach. Learn., 2018, vol. 80, pp. 2965–2974.
• [E. Dalsasso, 21a] E. Dalsasso, L. Denis, and F. Tupin, “SAR2SAR: A semi-
supervised despeckling algorithm for SAR images,” IEEE J. Sel. Topics Appl.
Earth Observ. Remote Sens., vol. 14, pp. 4321–4329, 2021.
• [E. Dalsasso, 21b] E. Dalsasso, L. Denis and F. Tupin, (2021), “As if by magic: self-supervised training of deep despeckling networks with MERLIN”, arXiv preprint arXiv:2110.13148.
• [I. S. Reed, 90] I. S. Reed and X. Yu, “Adaptive multiple-band CFAR detection of an optical pattern with unknown spectral distribution,” IEEE Transactions on acoustics, speech, and Signal Processing, vol. 38, no. 10, pp. 1760–1770, 1990.
• [M. Caron, 21] M. Caron, H. Touvron, I. Misra, H. Jégou, J. Mairal, P. Bo-
janowski, and A. Joulin. Emerging properties in self-supervised vision transformers, in Proceedings of the International Conference on Computer Vision (ICCV), 2021.
• [A. Mian, 19] A. Mian, J.-P. Ovarlez, A. M. Atto and G. Ginolhac, “Design of New Wavelet Packets Adapted to High-Resolution SAR Images With an Application to Target Detection”, Geoscience and Remote Sensing, IEEE
Transactions on, 57(6), pp.3919-3932, June 2019.
• [Pottier, 09] J.-S. Lee and E. Pottier, “Polarimetric Radar Imaging: From
Basics to Applications”, CRC Press, 2009.
• [Barrachina, 23] J.-A. Barrachina, C. Ren, G. Vieillard, C. Morisseau, and J.-
P. Ovarlez, “Theory and implementation of complex-valued neural networks,” arXiv preprint arXiv:2302.08286, Feb. 2023.

Profil du candidat :
Master in machine learning, applied mathematics, statistics, or signal processing. Good technical skills in programming. Eager to work in the radar and SAR imaging field.

Formation et compétences requises :
Master in machine learning, applied mathematics, statistics, or signal processing. Good technical skills in programming. Eager to work in the radar and SAR imaging field.

Adresse d’emploi :
The Ph.D. student will be hosted at the SONDRA laboratory (joint international laboratory between CentraleSupélec, ONERA, DSO National Laboratories, and National University of Singapore) in Paris-Saclay campus in Gif-sur-Yvette and at the MATS research unit (Advanced Methods in Signal Processing) of the Electromagnetism and Radar Department at ONERA’s Palaiseau site. Due to the international visibility of the lab, some overseas exchanges with Singapore could be easily considered. The SONDRA laboratory may finance any conference travel by the doctoral student.

Document attaché : 202312071051_Self_Supervised_Anomaly_Detection_in_complex_valued_SAR_imaging.pdf