JDSE Autumn 2021


The sixth edition of the Paris-Saclay Junior Conference on Data Science and Engineering (JDSE) is addressed to first-year Ph.D. students, M2 students, and third-year students at Engineering schools at Paris-Saclay and Institut Polytechnique de Paris. It will offer these students the opportunity to present the scientific work developed during their internships or the first year of their Ph.D. thesis. Also, it will allow them to exchange ideas and discuss ongoing research with prestigious invited speakers, academics, and industry scientists, as well as their peers.

The conference aims at gathering a vast audience and is an excellent means of discovering research activities in Data Science and Engineering.

The event is co-organized by the Labex DigiCosme and DATAIA

* Registration and the sanitary pass will be required to come on-site (within the limit of available seats).
Registration for meals will close on September 24, 2021.

Place

Amphi 1, Eiffel building, CentraleSupélec AND online

Program

Our keynote speakers :

  • Camille Couprie, Research scientist at Facebook Artificial Intelligence Research

Deep vectorized, surprising or realistic image generation
Deep image generation is becoming a tool to enhance artists and designers’ creativity potential. It is also becoming mature enough to potentially serve more classical applications, such as super-resolution and compression. This talk is structured into two main parts. In the first one, we introduce a self-supervised image decomposition approach in vector layers with applications in image editing, vectorization and retrieval. Then, we build on recent image retrieval, completion, and composition methods to design a new collage generation tool to assist artists in creating interesting composite images. In the second part, our purpose is on the contrary to generate realistic images and videos. In particular, we will see how to optimize latent code to improve a super-resolution approach. Finally, we introduce a video chat compression approach that transmit compressed facial landmarks, reconstructs accurate faces, and runs on mobile phones.

Bio: Camille Couprie is a research scientist at Facebook Artificial Intelligence Research. Previously she worked at IFP énergies nouvelles, a french research organization active in the fields of energy, transport and environment. In 2012, she was a postdoctoral fellow in Yann LeCun’s lab at the Courant Institute of Mathematical Sciences, New York University. Camille received her PhD in computer science from Université Paris Est / ESIEE Paris in 2011, under the supervision of Hugues Talbot, Laurent Najman and Leo Grady. She received two awards for best PhD from the EADS foundation and from the French Armament Agency, as well as an accessit Gilles Khan best PhD award.

  • Francis Bach, INRIA – SIERRA project-team, Département d’Informatique de l’Ecole Normale Supérieure, PSL Research University

“Convex and non-convex optimization for machine learning
Machine learning is naturally cast as an optimization problem, where the error rate on typically large-scale observed data is minimized. Designing efficient algorithms and providing associated guarantees is a crucial mathematical challenge which remains an active area of research owing to the breadth of machine learning models. In this talk, I will present the main optimization problems encountered in machine learning, from convex to non-convex, and highlight the associated key difficulties and open problems.

Francis Bach is a researcher at Inria, leading since 2011 the machine learning team which is part of the Computer Science department at Ecole Normale Supérieure. He graduated from Ecole Polytechnique in 1997 and completed his Ph.D. in Computer Science at U.C. Berkeley in 2005, working with Professor Michael Jordan. He spent two years in the Mathematical Morphology group at Ecole des Mines de Paris, then he joined the computer vision project-team at Inria/Ecole Normale Supérieure from 2007 to 2010. Francis Bach is primarily interested in machine learning, and especially in sparse methods, kernel-based learning, large-scale optimization, computer vision and signal processing. He obtained in 2009 a Starting Grant and in 2016 a Consolidator Grant from the European Research Council, and received the Inria young researcher prize in 2012, the ICML test-of-time award in 2014 and 2019, as well as the Lagrange prize in continuous optimization in 2018, and the Jean-Jacques Moreau prize in 2019. He was elected in 2020 at the French Academy of Sciences. In 2015, he was program co-chair of the International Conference in Machine learning (ICML), and general chair in 2018; he is now co-editor-in-chief of the Journal of Machine Learning Research.

  • Paul Poncet, ENGIE, Head of Data Science for the Darwin platform at Engie Digital

Predictive maintenance of renewable energy assets: what can data science really do?

We first give some background on renewable energy assets and the data we collect from sensors that measure mechanical, electrical, and meteorological information there. Then, going through a variety of concrete renewable energy cases around predictive maintenance and anomaly detection in time series, we reflect on both the kind of machine learning we need and the technical role a data scientist plays in this context, from the creation of an algorithm up to its industrialization.

Paul Poncet has 15 years of experience in the Energy industry; he is currently leading a team of data scientists and data engineers at Engie Digital who invent, develop, and deploy algorithms for the renewable energy businesses. He owns a Master degree in Engineering from Ecole des Mines de Paris, and a PhD in Applied Mathematics from Ecole Polytechnique.

  • Geoffroy Peeters, Télécom-Paris, Institut Polytechnique de Paris, LTCI, IDS, S2A (Signal-Statistique-Apprentissage) team

“Deep learning for music audio signal processing

As in many fields, deep neural networks have allowed important advances in the processing of musical audio signals. We first present the specificities of these signals and some elements of audio signal processing (as used in the traditional machine-learning approach. We then show how deep neural networks (in particular convolutional
neural networks) can be used to perform feature learning. We first recall the fundamental differences between 2D images and time/frequency representations. We then discuss the choice of input (spectrogram, CQT, or raw-waveform), the choice of convolutional filter shape, autoregressive neural models, and the different ways of injecting a priori knowledge (harmonicity, source/filter) into these networks. Finally, we illustrate the different learning paradigms used in the music audio domain: classification, encoder-decoder (source separation,
constraints on latent space), metric learning (triplet loss), and semi-supervised learning.

30th of September 2021

  • 8:45- 9:15 Coffee and registration
  • 9:15 – 9:45 Welcome message and opening
  • 9:45 – 10:45 Keynote Speaker: Paul Poncet, ENGIE
  • 10:45 – 12:30 Coffee, viennoiseries break
  • 10:45 – 12:30 Poster session
  • 12:30 – 13:30 Lunch Break
  • 13:40 – 14:40 Keynote Speaker: Francis Bach, INRIA, ENS, PSL
  • 14:50 – 15:15 Pause Gouter
  • 15:15 – 16:15 Keynote Speaker: Camille Couprie, Facebook Artificial Intelligence Research
  • 16:15 – 16:45 Presentation of the student mentoring session
  • 17:00 – 19:00 Student Mentoring Cocktail Session (Pablo Piantanida, Camille Couprie, Paul Poncet, Joe Raad, Gordana Draskovic)

1st of octobre 2021

  • 8:45 – 9:15 Coffee viennoiseries and registration
  • 9:15 – 10:15 Keynote Speaker Geoffroy Peeters, Télécom-Paris
  • 10:00 – 10:30 Coffee break
  • 10:30 – 11:15 : Presentations :
  1. ML-CI: Machine Learning Confidence Intervals for Covid-19 forecasts, Alice Lacan, LISN/ Inria / CNRS
  2. DriPP: Driven Point Processes to Model Stimuli Induced Patterns in M/EEG Signals, Cédric Allain, Université Paris-Saclay / Inria / CEA
  3. Reservoir computing for Alzheimer’s disease detection, Nickson Mwamsojo, Telecom SudParis/ IP Paris

  • 11:15 – 11:30 Coffee break
  • 11:30 – 12:15 Presentations :

  1. Semantic Matching with a Lightweight Positional Attention Module, Lihu Chen, LTCI / Telecom Paris/ IP Paris
  2. Securing Federated Learning against Backdoor Attacks, Eva Feillet, Université Paris-Saclay/ CEA
  3. A new solution to tackle latent space collapse in Learning-based Point Cloud Compression, Hai Dang Nguyen, CentraleSupélec / Université Paris-Saclay

  • 12:15 – 13:45 Lunch time + best presentation and poster award

Organizers

The organization committee manages the logistics and communication aspects of the conference.

  • Stéphane Lathuilière
    Co-chair Telecom Paris
  • Giuseppe Valenzise
    Co-chair L2S, CNRS, CentraleSupelec
  • Ariane Migault
    Conference Communication Leader, Digicosme
  • Benoit Monegier du sorbier

Junior Organizing committee

  • Armita Khajeh Nassiri, Ph.D. student LISN, Université Paris-Saclay