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

Laboratoire/Entreprise : IFP Energies nouvelles, Lyon
Durée : 3-4 months
Contact : laurent.duval@ifpen.fr
Date limite de publication : 2022-12-26

Contexte :
Le génie des procédés est un domaine d’ingénierie, qui s’intéresse à l’application de la chimie physique dans l’industrie avec pour but principal de transformer de la matière. Il vise à concevoir et gérer le fonctionnement d’un procédé à différentes transformations chimiques et physiques. Ce domaine s’appuie sur l’analyse régulière de différents capteurs, paramètres et quantités physiques (température, pression, concentration, etc.), permettant d’optimiser les points de fonctionnement et la qualité des produits en fonction des matières sources. Être capable de bâtir des modèles statistiques prédictifs ou explicatifs, liant les données mesurées ou calculées aux propriétés souhaitées, devient donc un enjeu important en termes d’efficacité des procédés. Depuis plusieurs années se développent des approches combinant connaissances métiers et méthodes de science des données pour améliorer la performance et la robustesse des modèles ; les défis sont nombreux : disponibilités des données, incertitudes, validité des modèles physiques, capacité des modèles off-line expérimentaux à fonctionner en conditions industrielles réelles on-line, etc. Ce sujet s’intéresse essentiellement à ce dernier aspect. En effet, les données industrielles ne disposent bien souvent que de mesures acquises à une fréquence assez lente, sans maîtrise entière du moment exact de l’acquisition, tandis que les données expérimentales produites sur unités pilotes peuvent être obtenues toutes les minutes, avec une bonne précision temporelle.

Sujet :
La ou le stagiaire sera affecté(e) au sein de la direction Expérimentation Procédés et travaillera en étroite collaboration avec la direction Sciences et Technologies du Numérique. La ou le stagiaire devra analyser les données sur un procédé identifié dans la continuité d’un stage de 2021 et devra étudier l’impact de l’échantillonnage sur la fiabilité et la robustesse de modèles d’imputation et de prédiction développés. Le coeur du travail consistera en la mise en place d’un workflow permettant de réduire la fréquence d’acquisition initiale de données pilotes pour l’approcher des mesures industrielles et de mesurer précisément la perte d’efficacité et les incertitudes induites, et à adapter les modèles en conséquence. Il est également attendu de pouvoir donner des préconisations de « bon échantillonnage » en fonction des dépendances temporelles entre les différentes variables mesurées.

Information: http://www.laurent-duval.eu/job-2022-internship-process-engineering-data-science-ifpen-sampling-robustness.html

Profil du candidat :
Ce sujet étant à l’interface de deux domaines, il s’adresse soit à des étudiants M1 ou M2 en sciences du numérique ayant un sens physique développé ou à des étudiants M1 ou M2 en génie des procédés ayant une appétence pour le machine-learning et la programmation.

Formation et compétences requises :
Statistiques, apprentissage, traitement de séries temporelles, génie des procédés

Adresse d’emploi :
France, IFP Energies nouvelles, Lyon (Solaize)

Document attaché : 202202052239_job-2022-internship-ifpen-machine-learning-robustness-process-sampling-english.pdf

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

Laboratoire/Entreprise : IFREMER Brest
Durée : 16 months
Contact : laurent.duval@ifpen.fr
Date limite de publication : 2022-12-26

Contexte :
Marine seismology has made tremendous technological advances in the past decades: data recorded at the seafloor by ocean bottom seismometers (OBSs) are becoming widely available (eg IRIS consortium). An OBS is a multicomponent instrument able to continuously record pressure and earth motion. There are two types of OBSs: short-period instruments for recording high-frequency motions, and long-period instruments for acquiring a wider range of motions (cf. INSU-IPGP national OBS facility). With both instruments, OBSs record a superposition of a broad variety of signals generated by solid earth, ocean wave, biologic, ship sources and noise. These signals can be very different in amplitude, duration and frequency content. They however also overlap in those domains, making them hard to isolate from each other. That is why OBS data cannot yet be fully exploited by the seismological community, as they require more advanced processing and identification techniques.

Sujet :
This postdoctoral position funded by the BRUIT-FM project primarily aims at developing signal processing and machine learning techniques to classify and separate signals recorded by OBSs and to enhance earthquake waveforms and microseismic noise. It devotes to a better exploitation of non-seismological signals for defining a seafloor soundscape. Hence the moniker ”Ocean Bottom Noise Shazam”, from the famous music retrieval/identification application

Information: http://www.laurent-duval.eu/job-2022-postdoc-ocean-shazam-bruit-fm-data-science-signal-processing-marine-seismology.html

Profil du candidat :
PhD with strong interest in spectral analysis, adaptive filtering, machine learning, data science with a taste of physics.

Formation et compétences requises :
Languages: C/C++, Python/Matlab or similar. Seismology is a plus

Adresse d’emploi :
France, IFREMER, Brest

Document attaché : 202202052227_KER_S_2022_SUBJ_PSD-Bruit-FM-Postdoc-Ocean-Bottom-Noise-Shazam.pdf

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

Laboratoire/Entreprise : CERI Systèmes Numériques – IMT Nord Europe
Durée : 5 mois
Contact : jerry.lonlac@imt-nord-europe.fr
Date limite de publication : 2022-03-31

Contexte :
Frequent Itemset Mining (FIM, for short) is an active part of data mining field and an important first step in data analysis. In the last decades, FIM has been applied in a broad range of applications such as e-commerce [4], e-learning [10], malware detection [3]. The application of FIM in a wide range of domains with a proliferation of different type of data has inspired the development of numerous other pattern-mining techniques. Recently, gradual itemsets [6, 2, 11, 7, 9] have then been proposed for analysing numerical data and different algorithms have been designed to automatically extract gradual itemsets from different data model [12, 13, 1, 5]. Gradual itemsets aroused great interest for extracting frequent complex co-variations between numerical attributes in a multitude of areas. However, in some real- world applications, data are subject to noise and measurement error. To date, the effect of noise on classical frequent gradual itemset mining algorithms has been not addressed.

Sujet :
The goal of this work is to propose a noise tolerant gradual itemset model, which unlike classical gradual itemsets [2, 8] tolerates a controlled fraction of errors on the extent of the gradual itemset. By allowing noise, the proposed models will generalize the level-wise enu- meration of different forms of frequent gradual itemsets [2, 12, 8, 7] that can be extracted from different types of complex numerical data but obscured by noise.

References
[1] Aymeric Cˆome and Jerry Lonlac. Extracting frequent (closed) seasonal gradual patterns using closed itemset mining. In IEEE International Conference on Tools with Artificial Intelligence, ICTAI, pages 1442–1448, 2021.
[2] Lisa Di-Jorio, Anne Laurent, and Maguelonne Teisseire. Mining frequent gradual item- sets from large databases. In IDA, pages 297–308, 2009.
[3] Yiheng Duan, Xiao Fu, Bin Luo, Ziqi Wang, Jin Shi, and Xiaojiang Du. Detective: Automatically identify and analyze malware processes in forensic scenarios via dlls. In ICC, pages 5691–5696, 2015.
[4] Philippe Fournier-Viger, Jerry Chun-Wei Lin, Bay Vo, Tin Chi Truong, Ji Zhang, and Hoai Bac Le. A survey of itemset mining. Wiley Interdiscip. Rev. Data Min. Knowl. Discov., 7(4), 2017.
[5] Amel Hidouri, Sa ̈ıd Jabbour, Jerry Lonlac, and Badran Raddaoui. A constraint-based approach for enumerating gradual itemsets. In IEEE International Conference on Tools with Artificial Intelligence, ICTAI, pages 582–589, 2021.
[6] Eyke Hu ̈llermeier. Association rules for expressing gradual dependencies. In PKDD, pages 200–211, 2002.
[7] Jerry Lonlac, Arnaud Doniec, Marin Lujak, and St ́ephane Lecoeuche. Mining frequent seasonal gradual patterns. In Big Data Analytics and Knowledge Discovery – DaWaK, volume 12393, pages 197–207, 2020.
[8] Jerry Lonlac, Yannick Miras, Aude Beauger, Vincent Mazenod, Jean-Luc Peiry, and Engelbert Mephu Nguifo. An approach for extracting frequent (closed) gradual patterns under temporal constraint. In FUZZ-IEEE, pages 878–885, 2018.
[9] Jerry Lonlac and Engelbert Mephu Nguifo. A novel algorithm for searching frequent gradual patterns from an ordered data set. Intell. Data Anal., 24(5):1029–1042, 2020.
[10] Esp ́erance Mwamikazi, Philippe Fournier-Viger, Chadia Moghrabi, and Robert Bau- douin. A dynamic questionnaire to further reduce questions in learning style assess- ment. In Lazaros Iliadis, Ilias Maglogiannis, and Harris Papadopoulos, editors, Artificial Intelligence Applications and Innovations, pages 224–235, 2014.
[11] Benjamin N ́egrevergne, Alexandre Termier, Marie-Christine Rousset, and Jean-Franc ̧ois M ́ehaut. Para miner: a generic pattern mining algorithm for multi-core architectures. DMKD, 28(3):593–633, 2014.
[12] NhatHai Phan, Dino Ienco, Donato Malerba, Pascal Poncelet, and Maguelonne Teis- seire. Mining multi-relational gradual patterns. In SDM, pages 846–854, 2015.
[13] Faaiz Shah, Arnaud Castelltort, and Anne Laurent. Extracting fuzzy gradual patterns from property graphs. In FUZZ-IEEE, pages 1–6, 2019.

Profil du candidat :
– 2nd year student of a Master’s or Engineering of Computer Science degree.
– Goods skills in Artificial Intelligence, more particularly in pattern mining.
– Goods skills in programming language (C++, Python).

Formation et compétences requises :
– 2nd year student of a Master’s or Engineering of Computer Science degree.
– Goods skills in Artificial Intelligence, more particularly in pattern mining.
– Goods skills in programming language (C++, Python).

Adresse d’emploi :
IMT Nord Europe
941, rue Charles Bourseul
CS 10838
59508 DOUAI Cedex – France

Document attaché : 202202041800_Proposal_for_internship_IMT_Nord_Europe_2022.pdf

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

Laboratoire/Entreprise : LIP6 – Sorbonne Université
Durée : 6 months
Contact : esteban.bautista-ruiz@lip6.fr
Date limite de publication : 2022-03-01

Contexte :
Link streams are sequences of interactions over time. They model a large number of datasets that have both a temporal and a structural component: phone calls, social interactions, internet traffic or financial transactions. The wealth of information contained in link streams bears great potential for progress in high-impact areas. For instance, frauds or thefts in monetary transactions may leave signatures expressed as substreams that heavily interact in a short span of time. Another example are network attacks which may be characterized as repetitive bursts of links that deviate from normal activity. Our goal in this internship is to develop algorithms that allow us to efficiently detect such anomalies in link streams.

Sujet :
The goal of this internship is to develop algorithms that can detect anomalies by ranking the importance of interactions in link streams. Recent works based on this idea have been used to detect microcluster anomalies: suddenly arriving groups of suspiciously similar links. However, such works rank interactions solely based on their time properties (how often two individuals interact) and overlook the structural properties of interactions. We therefore aim to develop ranking algorithms that take into account both the time and structural properties of link streams.

Numerous recent works have extended graph theory concepts to link streams, providing a solid foundation to analyze their structural properties. It is thus a timely challenge to leverage these concepts in the context of anomaly detection algorithms. We are particularly interested in exploring the recent definitions of temporal random walks, which have achieved great success in various applications. We are also open to explore other centrality metrics, such as betweenness, which raise important algorithmic challenges but have great interpretability.

Profil du candidat :
This internship is directed at M2 students with various background (complex networks, algorithmic, graph theory) with a strong interest in graph algorithmics and/or graph theory and its applications.

Formation et compétences requises :
Students in M2 with knowledge of the following: graph theory, algorithmique, statistics, statistical physics, and coding in Python and a compiled language like C/C++/Go/Rust.

Adresse d’emploi :
Le stagiaire fera partie de l’équipe Complex Networks du LIP6 – Sorbonne Université, situé à:
4 place Jussieu
75252 PARIS CEDEX 05, France

Document attaché : 202202011233_Internship_Proposal.pdf