Five new interns have just joined the DSAIDIS Chair: Yousef Tahiri Sojasi, Junjie Yang, Tamim El Ahmad, Oskar Rynkiewicz and Orson Jay. Below is a description of the topics they will be working on.

Tamim El Ahmad is in Year 2 of the Mathematics, Vision and Learning (MVA) Master’s program at ENS Paris-Saclay, a Master’s in mathematics, specialized in machine learning and computer vision.  He was previously in Year 1 of the Applied Mathematics Master’s program at Paris-Diderot and in the Master’s Degree in Engineering at Mines Saint-Etienne.

During his internship, his focus is on the development of hybrid architectures combining neural networks and kernel methods in order to solve prediction problems relating to structured data (especially link prediction, sequence and graph prediction). Florence d’Alché-Buc, professor at Télécom Paris and holder of the DSAIDIS Chair is supervising the internship.




Junjie Yang is in Year 2 of the École d’Ingénieurs ParisTech Shanghai Jiao Tong (SJTU-ParisTech) Master’s program. He has carried out research on automatic natural language processing, especially around such issues as question answering and natural language understanding via machine learning.

The aim of the internship is to apply several groups of methods to predict the structured output of text data, in particular the generation of text, which may have to perform a large number of potentially very different tasks (abstractive summarization, question answering, automatic image description, machine translation…). Matthieu Labeau, associate professor at Télécom Paris and Florence d’Alché-Buc are supervising the internship.



Oskar Rynkiewicz is an IASD Master’s student at Paris Dauphine-PSL university. He takes an interest in optimization and machine learning.

During his internship, he seeks to prove a lower complexity bound of primal-dual algorithms for convex affinely constrained optimization problems under metric sub-regularity. Once obtained, the lower bound will verify the optimality of the currently used methods with respect to metric sub-regularity. Olivier Fercoq, teacher-researcher at Télécom Paris, is supervising the internship.




Orson Jay is a student at Ecole Centrale de Nantes, in Year 1 of the Statistics and Data Science Master’s program.

During his internship, he is studying the general skills employers look for in asynchronous video interviews during his internship. He is using a corpus analysis method, studying recent literature on multi-label and label embedding models, and implementing recurrent neural networks based on LSTM or GRU units to solve multi-label problems. Chloé Clavel, teacher-researcher at Télécom Paris and Léo Hemamou, doctoral student at Télécom Paris are supervising the internship.



Yousef Taheri Sojasi is passionate about machine learning and natural language processing.

During his internship, he will be developing representation methods in order to make it easier to detect weak signals in text data. Weak signal detection is a major challenge when it comes to applications.  The method takes its inspiration from methods and criteria based on extreme value theory, which extend the scope of supervised and unsupervised learning techniques.




Halim Hizaoui is an engineer-student at ENSTA Paris on a double-diploma course, ENSA-ENIT (Ecole nationale d’ingénieurs de Tunis). He is passionate about finance and machine learning.

He will be developing a supervised classification method, inspired by the DDaplha classifier (R-package ddalpha). The DDalpha classifier is based on data depth and the alpha-procedure. It offers the benefits of a being non-parametric, fast and robust. The alpha-procedure is an iterative heuristic procedure which results in a discrete space of relevant variables. Function loss optimization takes less computing time thanks to its recursive nature.  The combination of data depth and the alpha-procedure provides promising results, which warrant further development of this method on the heuristics side. The internship will therefore focus on research on the alpha-procedure. Real and synthetic databases will be tested in order to establish the optimal parameters for each configuration and aid decision-making when selecting these parameters. Halim will be supervised by Pavlo Mozharovskyi, associate professor at Télécom Paris.