Archives

24 post doc position in deep learning for cancerology

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

Laboratoire/Entreprise : Sesstim, Aix Marseille University
Durée : 24 Mois
Contact : raquel.urena@univ-amu.fr
Date limite de publication : 2023-06-26

Contexte :
The growing life expectancy in developed countries has led to an alarming rise in cancer cases among the elderly. Age is a significant risk factor, with cancer incidence 10 times higher in patients over 65 years old. Sadly, limited clinical studies focus on elderly patients due to exclusion criteria, resulting in suboptimal treatment decisions. Embracing the power of AI, we aim to enhance early detection, diagnosis, and treatment prediction in oncology, particularly for the elderly population.

Sujet :
Join us in leveraging advanced Machine Learning and Deep Learning techniques to analyze clinical data and healthcare reimbursement records from the French Cancer Data Platform (INCA). Our objective is to construct comprehensive temporal profiles for patients, considering clinical characteristics, treatment history, comorbidities, and care utilization using cutting edge deep learning models such us AED and transformers. The study will specifically focus on patients aged 70 and above receiving care at the Institut Paoli-Calmette in Marseille. By developing predictive algorithms, we strive to improve treatment response predictions for elderly cancer patients, thus enhancing their medical care.
Research Team: Collaborate with the esteemed Sesstim researchers and oncologists at Marseille’s Cancer Hospital, IPC. Our multidisciplinary team consists of data scientists, medical professionals, and statisticians. As a postdoctoral researcher, you will receive direct supervision Raquel Urena,associate professor and AI specialist and A.D. Bouhnik biostatistician and specialist in Cancer research along with oncologists and public health doctors. Your primary location will be the Faculty of Medicine in Marseille.

Profil du candidat :

• Strong publication record in deep learning and machine learning
• Proficiency in Python and R programming
• In-depth knowledge of machine learning, deep learning methodologies and LLM
• Solid understanding of SQL databases
• Fluent in French and English

Formation et compétences requises :
-PhD in Artificial Intelligence or Computer Science
-MS degree in Computer Science, Statistics, Maths or related discipline

Adresse d’emploi :
Université Aix-Marseille – Faculté de Médecine – 27, boulevard Jean Moulin 13385 Marseille Cedex 5

Document attaché : 202306230927_24 months Post Doc position in artificial intelligence applied to cancerology research.pdf

Data Privacy on Graphs with semantic information

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

Laboratoire/Entreprise : INSA Centre Val-de-Loire, Laboratoire LIFO
Durée : 3 ans
Contact : adrien.boiret@insa-cvl.fr
Date limite de publication : 2023-12-31

Contexte :
The Systems and Data Security Team of Laboratoire d’Informatique Fondamentale d’Orléans is offering a PhD position to study data privacy in graph databases containing semantic information.
This PhD position is part of the CyberINSA project.

Sujet :
We aim to explore explore how privacy guaranties potentially weaken in the case where graph databases respect a known schema or ontology, and to present adaptations and countermeasures.
Full subject in the attached file.

Profil du candidat :
• Research Master in computer science / engineering
• Knowledge or interest about databases (especially graph databases, e.g.
RDF) and data privacy

Formation et compétences requises :
• Ability to read and write English documents
• Proficiency in a coding language is preferred
• Willingness to work in autonomy and in a team

Adresse d’emploi :
INSA Centre Val de Loire, Bourges Campus
88 Boulevard Lahitolle, 18000 Bourges

Document attaché : 202306202242_These-Final.pdf

EDA 2023

Date : 2023-10-26 => 2023-10-27
Lieu : Montpellier, France

EDA 2023 : 19èmes journées Business Intelligence & Big Data

Edition conjointe avec BDA 2023
Montpellier, France, 26-27 Octobre 2023

Site web de la conférence : https://eda2023.sciencesconf.org/

Lien pour les soumissions : https://eda2023.sciencesconf.org/

Date limite de soumission : 16 JUILLET 2023

L’objectif principal des Journées EDA est de fournir un forum pour la dissémination des réalisations de la recherche et de promouvoir des interactions et des collaborations dans le domaine des systèmes d’information décisionnels (Business Intelligence) et l’analyse des mégadonnées (big data analytics). Ces deux domaines poussent encore plus loin les défis scientifiques, par des données toujours plus multiples et complexes (big queries, big times series, …) dans des environnements de stockage toujours plus sophistiqués (les data lakes, en particulier) et tout en devant assurer une bonne compréhension des analyses possibles. Les 19èmes journées EDA “Business Intelligence & Big Data” se tiendront intégralement à Montpellier les 26 et 27 Octobre 2023. Nous invitons les chercheurs, les doctorants et les experts intéressés par ces domaines à présenter leurs travaux de recherche ou des applications développées autour d’originalités technologiques.

Cette édition se tiendra conjointement avec la 39ème conférence BDA « Gestion de Données – Principes, Technologies et Applications » (du 23 au 26 octobre 2023).

Lien direct

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

MCF CDD à CY Cergy Paris Université (CYU)

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

Laboratoire/Entreprise : ETIS / CY Cergy Paris Université
Durée : 1 an renouvelable
Contact : dan.vodislav@u-cergy.fr
Date limite de publication : 2023-07-21

Contexte :
Ce poste est donc lié aux Bachelors internationaux « Data Science and Big Data Technology » en collaboration avec la Zhejiang University of Science and Technology (ZUST), à Hangzhou, en Chine, et « Data Science » en collaboration avec l’Université de Maurice (UoM). Le service d’enseignement sera partagé entre les deux Bachelor et d’autres enseignements au sein du département de sciences informatiques de CYU. La recherche se déroulera au sein du laboratoire ETIS.
Pour candidater, envoyer CV et lettre de motivation à dan.vodislav@cyu.fr et dimitrios.kotzinos@cyu.fr.
Date limite de candidature: 5 juillet 2023

Sujet :
Le poste est ouvert sous la forme d’un CDD initial d’un an, à partir de septembre 2023, avec la volonté de proposer par la suite une extension avec un contrat de 3 ans.
Dans le cadre des deux Bachelors, l’enseignant recruté participera à l’enseignement et au pilotage pédagogique des modules de Bachelor, au sein d’une équipe pédagogique de plusieurs enseignants et enseignants-chercheurs. L’enseignement dans les deux Bachelors se fait sur place, en Chine et à Maurice, lors de séjours de quelques semaines, avec plusieurs séjours à prévoir par an. Pour le Bachelor avec ZUST l’enseignement se fait en français, pour celui avec l’UoM l’enseignement se fait en anglais.
En recherche, l’intégration au laboratoire ETIS se fera possiblement dans l’équipe MIDI, sur des thématiques de recherche autour de l’intégration et l’analyse de grandes masses de données de divers types, s’appuyant notamment sur l’élaboration de méthodes d’apprentissage automatique pour l’analyse et la gestion de ces données.

Profil du candidat :
Voir la fiche de poste sur le site https://cytech.cyu.fr/cytech/institut-sciences-et-techniques, rubrique Recrutements

Formation et compétences requises :
Titulaire d’un doctorat en informatique, de préférence en lien avec la gestion de données, avec une expérience d’enseignement en Licence et Master en informatique.
Mobilité internationale requise.

Adresse d’emploi :
CY Cergy Paris Université
2 avenue Adolphe Chauvin
95000 Pontoise

Méthodes d’apprentissage profonds visant le contrôle des structures critiques : vers des solutions en quasi temps réel pour la résolution de problèmes directes et inverses

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

Laboratoire/Entreprise : CEA List
Durée : 36 mois
Contact : roberto.miorelli@cea.fr
Date limite de publication : 2023-07-21

Contexte :
In this thesis, we focus on the study and the application of generative methods aiming at enhancing the quality of predictions in case of forward and inverse ML models (i.e., metamodels) for safety critical problems in industry.

Sujet :
L’utilisation des méthodes d’Intelligence Artificielle (i.e., méthodes d’apprentissage par réseaux neuronaux profondes) ouvre des perspectives très intéressantes dans le domaine du contrôle non destructif (CND) appliqué dans le domaine industriel comme par exemple l’assistance au diagnostic sur la base de mesures non-destructives. Récemment, l’utilisation des techniques d’apprentissage profond a montré des premières résultats encourageants dans le domaine du contrôle de la santé intégré [1], dans le contrôle des températures par méthodes d’imagerie infrarouge dans les centrales de fusion nucléaires [2, 3] et pour la génération de données réalistes dans le cas d’imagerie par ultrasons, entre autres.

Néanmoins, malgré plusieurs succès récents des techniques d’apprentissage automatique appliquées au CND, des verrous restent encore à lever pour rendre performantes et robustes ces techniques et les utiliser de manière fiable et systématique dans le domaine du CND hautement critique. Cette thèse vise à répondre à deux principaux points qui jouent un rôle dominant sur les performances des modèles d’apprentissage : i) le manque de données labélisées disponibles pour l’apprentissage des réseaux de neurones et ii) l’impact de incertitudes, qu’elles soient aléatoires ou bien épistémiques. Dans ce contexte, l’usage de la simulation permet de dépasser une grande partie des limitations actuelles, liées au manque de données de terrain labellisées, en créant de grandes bases de données synthétiques d’apprentissage, capable de couvrir tous types de scénarios et des cas non encore observés. La difficulté, cependant ici, est d’être capable de gérer les incertitudes inhérentes à la mesure expérimentale (erreur de calibration, bruit de mesure, etc.) et les incertitudes du modèle lui-même (i.e., les erreurs à la fois du modèle physique utilisé pour l’apprentissage et celle du modèle d’apprentissage –réseaux de neurones).

Cette thèse vise à améliorer la qualité des prédictions par IA dans le cas des modèles direct (de l’observable à la mesure) et dans le cas de modèles inverses (de la mesure à l’observable). En première lieu, une attention particulière sera mise sur la conception d’outils d’apprentissage profond de type générative conditionnées (e.g., auto-encodeurs variationels conditionnés, architectures de type UNET conditionnées, etc.) -reposant sur l’utilisation de données multi-fidélités- pour la génération de données réalistes, l’analyse et l’optimisation des problèmes d’inspection CND. Dans une deuxième étape, une forte attention sera donnée aux schémas d’apprentissage profond capable de promouvoir l’estimation des incertitudes (e.g., méthodes d’ensemble, Monte Carlo drop out, etc.) associées à la tâche d’apprentissage menées (i.e., régression, classification, etc.).

Dans ce travail de thèse, l’application des outils d’apprentissage développées se fera dans deux principaux domaines d’intérêt : le contrôle des parois des centrales de fusion nucléaire par imagerie infrarouge et l’inspection de pièces industrielles par imagerie ultrasonore.

———————————————————————————————-
English version:
———————————————————————————————-

Title: Enhancing the monitoring capabilities for safety critical problems via deep learning-based models: toward quasi real-time forward and inverse applications

Machine learning (ML) and in particular deep learning (DL) methods are gaining the attention of the engineering scientific and industry communities for enhance data analysis capabilities, in supporting human decisions, etc.. In the context of nondestructive testing and evaluation (NDT&E) and structural health monitoring (SHM), the use of ML methods is gaining the attention of scholars, researchers and experts. Indeed, the possibility to develop and deploy tailored ML strategies for detecting, classify and possibly provide quantitative information on anomalies (i.e., defects) in inspected structures in one of the most active investigation field in the community. That is, the application of deep learning methods (i.e., deep convolutional neural networks) has been recently applied with success in the domain of SHM based on guided wave imaging data [1], in the field of infrared thermography applied to tokamak plasma temperature monitoring [2, 3] and for the realistic simulation of ultrasound testing images under uncertainties, for instance.

Nevertheless, despite some recent successes in applying ML schema in NDT&E problems, there is still room of improvements different directions. Indeed, in NDT&E one faces two main challenges i) the chronical lack of properly labelled experimental data (e.g., security and secrecy issues) and ii) the impact of uncertainties on the measurements (e.g., experimental conditions, knowledge on the actual material properties). In this context, the use of advanced numerical models can be used to mitigate the impact of such issues by exploiting simulations results to be integrated into the ML model (the so-called model-driven ML) and possibly coupled to experimental data too.

In this thesis, we focus on the study and the application of generative methods aiming at enhancing the quality of predictions in case of forward and inverse ML models (i.e., metamodels). Firstly, a particular emphasis will be given on deep learning schemas aiming at enhancing the computational performance of advanced numerical solvers. Toward this end, conditional generative models (e.g., cVAE, cUNET, etc.) based on multi-fidelity data will be considered for fast and reliable generation of data for understanding, analyzing and optimize the NDT&E scenario considered. In a second and tightly related stage, a particular focus will be given on the study of deep learning strategies (e.g., deep ensembles, Monte Carlo drop out, etc.) aiming at performing forward and inverse tasks providing the uncertainty estimation associated to the predictions.

In the context of this thesis the use of data issued from infrared thermography and ultrasound testing will be privileged with a specific emphasis on imaging post-processed data.

Ref.:

[1] Miorelli, Roberto, et al. “Defect sizing in guided wave imaging structural health monitoring using convolutional neural networks.” NDT & E International 122 (2021): 102480.

[2] Juven et al., “Temperature Estimation in Fusion Devices using Machine Learning techniques on Infrared Specular Synthetic Data,” 2022 IEEE 14th Image, Video, and Multidimensional Signal Processing Workshop (IVMSP), 2022, pp. 1-5, doi: 10.1109/IVMSP54334.2022.9816270.

[3] Aumeunier, M. H., et al. “Development of inverse methods for infrared thermography in fusion devices.” Nuclear Materials and Energy 33 (2022): 101231.

Contacts : Roberto Miorelli, Ph.D.

Université Paris-Saclay, CEA, List – Département Instrumentation Numérique

roberto.miorelli@cea.fr;

Profil du candidat :
Niveau M2 en Physique, Mathématiques Appliquées ou Statistique

Formation et compétences requises :
Niveau M2 en Physique, Mathématiques Appliquées ou Statistique

Adresse d’emploi :
CEA Saclay

Atelier “Éthique dans l’Enseignement Supérieur et la Recherche : Occasions et Capacités”

Date : 2023-07-04
Lieu : Centre Inria de l’Université de Rennes (retransmission en visio)

À l’ère du financement par projet, de la mise en concurrence des chercheurs et chercheuses, des politiques productivistes dans le monde de la recherche, des partenariats public-privé, du transfert technologique, le tout dans un contexte de crise économique, écologique et sociale, il est urgent de mener des réflexions sur le rôle éthique des travailleurs et travailleuses de l’enseignement supérieur et de la recherche.

L’association de doctorant-e-s Nicomaque organise le 4 juillet 2023 de 15h à 18h un atelier portant sur ces thématiques au centre Inria de l’Université de Rennes.

Au travers de cet atelier, nous soulèverons les problématiques suivantes :

  • Quelles responsabilités portent l’Enseignement Supérieur et la Recherche (ESR) dans la réflexion éthique ?
  • Quelles sont les occasions qui nous poussent à la réflexion éthique dans notre travail au quotidien ?
  • Quelles sont nos capacités d’actions sur notre travail de recherche et d’enseignement ?

Au cours de cet atelier interviendront :

  • Enka Blanchard, chargée de recherche au CNRS, chercheuse transdisciplinaire, ayant travaillé sur les thématiques des politiques productivistes dans la recherche et autrice de travaux technocritiques sur les technologies du numérique
  • Bernard Friot, professeur émérite, sociologue et économiste, membre de Réseau Salariat, connu pour son travail sur le statut de la fonction publique, la sécurité sociale et le salaire rattaché à la personne

    • L’évènement est ouvert à tou-te-s sur réservation (voir lien ci-dessous), et sera retransmis en visioconférence.

    Plus d’info ici : https://atelier-ethique-inria2023.gitlab.io/

    Lien direct

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

CIFRE PhD : Building a Large Patients Graph Data Lake for Primary Care Medicine

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

Laboratoire/Entreprise : Télécom SudParis / Aldebaran
Durée : 36 mois
Contact : Amel.Bouzeghoub@telecom-sudparis.eu
Date limite de publication : 2023-07-07

Contexte :
This thesis is a CIFRE and a collaboration between Telecom SudParis and Aldebaran. The position will start in October 2023.

Keywords: Knowledge Graphs, Graph Database, Medical Ontology, Medical History taking

Sujet :
The aim of this PhD thesis is to build a graph-oriented data lake to store all patient health data connected to a relevant medical ontology. Patient data are acquired through Aldebaran (https://aldebaran.care/) digital medical history-taking device that generates a medical report for generalist practitioners. The logic of medical history-taking is supported by a knowledge graph database comprising up to 600 questions, 3000 responses, and 6000 relationships. As of April 1st 2023, 2500 questionnaires and their medical reports have been generated. Of note, each response
of a patient to a question generates lexical annotations for the medical reports and semantic annotations supported by Conceptual Graphs (CG). CGs are the basis of the hyper-contextuality of the medical history tracking device. As it supports the semantics of the response, it will also populate the patient data graph. The main objectives of this thesis project are threefold : (i) to propose a model for patients’ reported health needs, characteristics, and continuum of
care using knowledge graphs and including existing standards (OMOP, OHDSI, HL7-FIHR) in a future perspective of interoperability and data analysis ; (ii) to provide a safe and efficient mean to enrich the patient file with longitudinal data acquired along successive medical consultations. In this respect, temporality management is a key challenge ; (iii) to propose a method to ensure the contextualization and conciseness of the questionnaire according to the elements already known in the patient data graph.

Profil du candidat :
For this thesis, we will consider candidates with a Master’s or Engineer ‘s degree with knowledge about several of the following skills:

– Fluent in written and spoken English. Some knowledge of French can be useful.
– Skills in mathematics (statistics, graphs) and computer science (algorithms, machine learning, knowledge, data modeling, symbolic artificial intelligence, Natural Language Processing)
– Mastery of Python language, Cypher, (React), and experience in software development
– Adaptability and ability to invest in the field of medical application
– Experience in a research laboratory

This thesis is a CIFRE and a collaboration between Telecom SudParis and Aldebaran. The position will start in October 2023.

Formation et compétences requises :
Master’s or engineer’s degree in Computer Science with an affinity for Machine Learning.

Adresse d’emploi :
The PhD student will be co-hosted by Telecom SudParis (Palaiseau site) and Aldebaran (Paris).

Applications should be submitted by email to Amel.Bouzeghoub@telecom-sudparis.eu, Julien.Romero@telecom-sudparis.eu, and christian@aldebaran.care

They must include the following:
– A Curriculum Vitae;
– Transcripts of records of undergraduate and graduate studies;
– Link to MSc thesis and publications if applicable;
– Link to personal software repositories
– Name of 2 or 3 references to contact (position, email);

post-doc position: machine learning for landslide (glissement de terrain) detection from interferometric SAR radar imagery

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

Laboratoire/Entreprise : ISTerre, Université Grenoble Alpes
Durée : 12 months
Contact : sophie.giffard@univ-grenoble-alpes.fr
Date limite de publication : 2023-01-09

Contexte :
In the mountains, slope movements such as landslides, Deep-seated Gravitational Slope Deformation (DGSD) or rock glaciers, manifest themselves through surface velocities ranging from a few mm/year to a few m/year, and can accelerate to catastrophic failure.
During their slow-moving phase, they can be detected in SAR interferograms (INSAR) acquired by the Sentinel-1 open-source global satellites. However, this detection is not automatic, despite the societal importance of such detection. However, Sentinel-1 InSAR data are now mass-processed at regional scales, by SNO ISDeform, making it possible to
automate this detection. Based on promising initial work and an initial manual inventory of the Alps currently under construction at ISTerre, this project aims to build a detection and classification algorithm for gravity instabilities in mountains, using a machine-learning algorithm for image segmentation (of the U-net convolutional neural network type).

Sujet :
As the inventory is already being built up by a current ISTerre post-doc, the person recruited will work in tandem with him to develop learning algorithms capable of detecting and characterizing slow landslides. The first part of the post-doc will be dedicated to getting to grips with the subject, the data and the literature. The main task will then be to develop a
specific algorithm taking into account the particular aspects of INSAR data (acquisition in radar geometry, complex image values, different durations between acquisitions) in order to make the most of the information contained in these data. Depending on the skills and progress of the person recruited, a number of perspectives may be considered: coupling
with optical data, analysis on different time scales, applications to other regions (South America). The person recruited will also be involved in the creation of an open source signal database that can be fed collaboratively.

Profil du candidat :
The profile sought is clearly linked to data sciences (particularly remote sensing) and techniques developed in artificial intelligence, with experience in methodological developments applied to geosciences.

Formation et compétences requises :
Experience in Python, cluster computing and remote sensing is required. Support in remote sensing and earth sciences, as well as in machine learning, will be available within the research team. A strong interest in the applicative aspects of methodological developments in AI and
curiosity about important processes in the earth sciences will be appreciated. As the working environment is interdisciplinary, communication and leadership skills will be required. A strong link must be forged with the Gricad computing center. Partial supervision of internships/theses on related subjects may be offered.

Adresse d’emploi :
ISTerre, Université Grenoble Alpes, France.

Contact sophie.giffard@univ-grenoble-alpes.fr and pascal.lacroix@univ-grenoble-alpes.fr with your CV and a brief explanation of why you are applying

Fine-grained, multimodal speech anonymization

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

Laboratoire/Entreprise : Inria Nancy & Lille
Durée : 36 mois
Contact : emmanuel.vincent@inria.fr
Date limite de publication : 2023-07-02

Contexte :
This PhD is part of the “Personal data protection” project of PEPR Cybersécurité, which aims to advance privacy preservation technology for various application sectors. It will be co-supervised by Emmanuel Vincent and Marc Tommasi. The PhD student will have the opportunity to spend time in both the Multispeech and Magnet teams, to collaborate with 9 other research teams in France and with the French data protection authority CNIL, and to contribute to the project’s overall goals including the organization of an anonymization challenge.

Sujet :
Large-scale collection, storage, and processing of speech data poses severe privacy threats [1]. Indeed, speech encapsulates a wealth of personal data (e.g., age and gender, ethnic origin, personality traits, health and socio-economic status, etc.) which can be linked to the speaker’s identity via metadata or via automatic speaker recognition. Speech data may also be used for voice spoofing using voice cloning software. With firm backing by privacy legislations such as the European general data protection regulation (GDPR), several initiatives are emerging to develop and evaluate privacy preservation solutions for speech technology. These include voice anonymization methods [2] which aim to conceal the speaker’s voice identity without degrading the utility for downstream tasks, and speaker re-identification attacks [3] which aim to assess the resulting privacy guarantees, e.g., in the scope of the VoicePrivacy challenge series [4].

The first objective of this PhD is to improve the privacy-utility tradeoff by better disentangling speaker identity from other attributes, and better decorrelating the underlying dimensions. Solutions may rely on suitable generative or self-supervised models [5, 6] or on adversarial learning [7]. The resulting privacy guarantees will be evaluated via stronger attackers, e.g., taking metadata into account.

The second objective is to extend the proposed audio-only approach to multimodal speech (audio, facial video, and gestures). Solutions will exploit existing facial anonymization technology [8]. A key difficulty will be to preserve the correlations between modalities, which are essential for training multimodal voice processing systems.

Depending on the PhD student’s skills, additional directions may also be explored, e.g., evaluating the proposed anonymization solutions in the context of federated learning.

[1] A. Nautsch, A. Jimenez, A. Treiber, J. Kolberg, C. Jasserand, E. Kindt, H. Delgado, M. Todisco, M. A. Hmani, M. A. Mtibaa, A. Abdelraheem, A. Abad, F. Teixeira, M. Gomez-Barrero, D. Petrovska, N. Chollet, G. Evans, T. Schneider, J.-F. Bonastre, B. Raj, I. Trancoso, and C. Busch, “Preserving privacy in speaker and speech characterisation,” Computer Speech and Language, vol. 58, pp. 441–480, 2019.

[2] B. M. L. Srivastava, M. Maouche, M. Sahidullah, E. Vincent, A. Bellet, M. Tommasi, N. Tomashenko, X. Wang, and J. Yamagishi, “Privacy and utility of x-vector based speaker anonymization,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, to appear.

[3] B. M. L. Srivastava, N. Vauquier, M. Sahidullah, A. Bellet, M. Tommasi, and E. Vincent, “Evaluating voice conversion-based privacy protection against informed attackers,” in 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 2802–2806, 2020.

[4] N. Tomashenko, X. Wang, E. Vincent, J. Patino, B. M. L. Srivastava, P.-G. Noé, A. Nautsch, N. Evans, J. Yamagishi, B. O’Brien, A. Chanclu, J.-F. Bonastre, M. Todisco, and M. Maouche, “The VoicePrivacy 2020 Challenge: Results and findings,” Computer Speech and Language, vol. 74, pp. 101362, 2022.

[5] L. Girin, S. Leglaive, X. Bie, J. Diard, T. Hueber, and X. Alameda-Pineda, “Dynamical variational autoencoders: A comprehensive review,” Now Foundations and Trends, 2021.

[6] A. Baevski, H. Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representations,” in Advances in Neural Information Processing Systems, pp. 12449–12460, 2020.

[7] B. M. L. Srivastava, A. Bellet, M. Tommasi, and E. Vincent, “Privacy-preserving adversarial representation learning in ASR: Reality or illusion?” in Interspeech, pp. 3700–3704, 2019.

[8] T. Ma, D. Li, W. Wang, and J. Dong, “CFA-Net: Controllable face anonymization network with identity representation manipulation,” arXiv preprint arXiv:2105.11137, 2021.

Profil du candidat :
MSc in computer science, machine learning, or signal processing.

Formation et compétences requises :
Strong programming skills in Python/Pytorch.
Prior experience in speech and video processing will be an asset.

Apply online at: https://jobs.inria.fr/public/classic/fr/offres/2023-06410

Adresse d’emploi :
615 Rue du Jardin-Botanique, 54600 Villers-lès-Nancy

Physics-Aware Deep Learning for Modeling Spatio-Temporal Dynamics.

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

Laboratoire/Entreprise : Sorbonne Universite – ISIR – Institut des Systèmes
Durée : 36 mois
Contact : patrick.gallinari@sorbonne-universite.fr
Date limite de publication : 2023-12-15

Contexte :
Physics-aware deep learning is an emerging research field aiming at investigating the potential of AI methods to advance scientific research for the modeling of complex natural phenomena. This research topic investigates how to leverage prior knowledge of first principles (physics) together with the ability of machine learning at extracting information from data. This is a fast-growing field with the potential to boost scientific progress and to change the way we develop research in a whole range of scientific domains. An area where this idea raises high hopes is the modeling of complex dynamics characterizing natural phenomena occurring in domains as diverse as climate science, earth science, biology, fluid dynamics, etc.

Sujet :
The objective of the PhD project is the development of Physics-aware deep learning methods for the modeling of complex spatio-temporal dynamics. The direct application of state-of-the-art deep learning (DL) methods for modeling and solving physical dynamics occurring in nature is limited by the complexity of the underlying phenomena, the need for large amounts of data and their inability to learn physically consistent laws. This has motivated the recent exploration of physics-aware methods incorporating prior physical knowledge. Although promising and rapidly developing, this research field faces several challenges. For this PhD project we will address two main challenges, namely the construction of hybrid models for integrating physics with DL and generalization issues which condition the usability of DL for physics.

— Integrating DL and physics for spatio-temporal dynamics forecasting and solving PDEs

In physics and many related fields, partial differential equations (PDEs) are the main tool for modeling and characterizing the dynamics underlying complex phenomena. Combining PDE models with ML is a natural idea when building physics-aware DL models and it is one of the key challenges in the field. This has been explored for two main directions: (i) augmenting low resolution solvers with ML in order to reach the accuracy of high-fidelity models at a reduced computational cost, and (ii) complementing incomplete physical models with ML by integrating observation data through machine learning. A first direction of the PhD will then be to investigate hybrid physics-DL models using the recently proposed framework of neural operators. The latter opens the possibility of combining and learning multiple spatio-temporal scales within a unified formalism, a challenge in DL.

— Domain generalization for deep learning based dynamical models

Explicit physical models come with guarantees and can be used in any context (also called domain or environment) where the model is valid. These models reflect explicit causality relations between the different variables involved in the model. This is not the case for DL: statistical models learn correlations from sample observations, their validity is usually limited to the context of the training domain. This is a critical issue for the adoption of ML for modeling the physical world. In relation with the construction of hybrid models as described above, one will investigate this issue along two main directions. The first one is a purely data-based approach and exploits ideas from learning from multiple environments through task decomposition. The second one, takes a dual perspective, relying on prior physical knowledge of the system equations and directly targets the problem of solving parametric PDEs, exploiting ideas from meta-learning.

Profil du candidat :
Computer science or applied mathematics. Good programming skills.

Formation et compétences requises :
Master degree in computer science or applied mathematics, Engineering school. Background and experience in machine learning.

Adresse d’emploi :
Sorbonne Université (S.U.), Pierre et Marie Campus in the center of Paris. The candidate will integrate the MLIA team (Machine Learning and Deep Learning for Information Access) at ISIR (Institut des Systèmes Intelligents et de Robotique).

Document attaché : 202306130923_2023-04-PhD-Description-Physics-Aware-Deep-Learning.pdf