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

Laboratoire/Entreprise : LITIS Lab (Rouen)
Durée : 5 to 6 months
Contact : paul.honeine@univ-rouen.fr
Date limite de publication : 2024-04-01

Contexte :
Safe and trustworthy Artificial Intelligence (AI) is central in the deployment of any AI system in major application areas, such as medicine and autonomous vehicles. Its major keystone requirements in Machine Learning (ML) have been recently investigated by researchers of the ML group in the LITIS Lab, including robustness, explainability and fairness. The current internship aims to address anomaly detection, which is a major ingredient of robust ML for Safe and trustworthy AI.

Sujet :
Self-supervised learning has recently emerged as a novel paradigm in Machine Learning, aiming to learn deep representations from unlabeled data. Its main driving force is contrastive self-supervised learning. A main ingredient in contrastive learning is a training scheme that contrasts each sample with augmented versions of itself, where augmentation strategies in imagery include color jittering, image rotation, image flipping and affine geometric transformations. Contrastive learning has been largely investigated for classification tasks, often demonstrating its relevance on well-known image classification benchmarks. However, such classification tasks with labelled training data do not get the most out of the self-supervised learning paradigm.

The goal of this internship is to explore contrastive learning for out-of-distribution detection in time series data. This would allow to take full advantage of the self-supervised learning paradigm for out-of-distribution detection (also called anomaly or novelty detection). The tasks to be carried out by the intern are as follows: The intern will implement different contrastive learning models. She/he will study augmentation methods that are relevant for time series, either by revisiting image transformations in the light of time series or by using distribution-shifting augmentations. The intern will conduct experiments on real time series by considering two contexts: detection from a batch of time series data, and online detection, namely in the context of streaming data.

This internship may lead to a PhD thesis.

Research Environment: This intern will conduct her/his research within the Machine Learning group in the LITIS Lab, under the supervision of Prof. Paul Honeine. This internship will be conducted within a research project gathering 9 permanent researchers of the LITIS Lab and the intern will also interact with several PhD students and interns also working on deep anomaly detection for time series.

References

– Hendrycks, Dan, Mantas Mazeika, Saurav Kadavath, and Dawn Song. “Using self-supervised learning can improve model robustness and uncertainty.” Advances in neural information processing systems 32 (2019).
– Li, Chun-Liang, Kihyuk Sohn, Jinsung Yoon, and Tomas Pfister. “Cutpaste: Self-supervised learning for anomaly detection and localization.” In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 9664-9674. 2021.
– Liu, Xiao, Fanjin Zhang, Zhenyu Hou, Li Mian, Zhaoyu Wang, Jing Zhang, and Jie Tang. “Self-supervised learning: Generative or contrastive.” IEEE transactions on knowledge and data engineering 35, no. 1 (2021): 857-876.
– Tack, Jihoon, Sangwoo Mo, Jongheon Jeong, and Jinwoo Shin. “CSI: Novelty detection via contrastive learning on distributionally shifted instances.” Advances in neural information processing systems 33 (2020): 11839-11852.

Profil du candidat :
Student in final year of Master or Engineering School, in data science, artificial intelligence, applied mathematics, or related fields.

Formation et compétences requises :
– Strong skills in advanced statistics and Machine Learning, including Deep Learning
– Good programming experience in Python

Adresse d’emploi :
LITIS Lab, University of Rouen Normandy, Saint Etienne du Rouvray (Rouen, France).

Application: Applicants are invited to send their CV and grade transcripts by email to paul.honeine@univ-rouen.fr.