Softskills Seminar

Content

Every student chooses one research paper from the list of proposed papers. The student then prepares a 30min presentation about this paper. For this purpose, she/he can request the help of the advisor of the paper (by email and/or by meeting with them). The student then gives the presentation in the allocated time slot of the Softskills seminar, in the presence of the lecturer. Students are warmly encouraged to take into account the advice on giving good talks dispensed during the first session.

Each presentation is followed by a question-answer session, where both the students and the lecturers can ask the presenter questions about the paper. To animate this, each student is assigned to some other paper as the “devil’s advocate”. In this role (which is not known to the other students), she or he prepares some questions for the presenter. However, all students are invited to participate in the question-answer session.

All participants should have received an email assigning them to one of the papers as a "devil".

Grading

• 80% presentation
• 20% oral participation (as devil’s advocate and in general).

Schedule

2023-09-26

1. Introduction
2. How to give good talks
3. How to do a PhD

2023-10-03 (does not take place)

2023-10-10

Julien Alexandre dit Sandretto 2: RUNGE–KUTTA THEORY AND CONSTRAINT PROGRAMMING (Antoine Stark)
Sophie Chabridon 1: I Prefer not to Say: Protecting User Consent in Models with Optional Personal Data (Bérénice Jaulmes)
Maria Boritchev 1: On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? (Wen YANG)

2023-10-17

Tiphaine Viard 1: Fairness and Abstraction in Sociotechnical Systems (Nicoline Nymand-Andersen)
Pietro Gori 1: A Simple Framework for Contrastive Learning of Visual Representations (Saranga MAHANTA)
Fabian Suchanek 2: Model Agnostic Supervised Local Explanations (Cody Clop)

2023-10-24

Matthieu Labeau 1: Unsupervised Detection of Contextualized Embedding Bias with Application to Ideology (Asma Khalil)
Matthieu Labeau 2: Efficient Hierarchical Domain Adaptation for Pretrained Language Models (Sheng-Yen LIN)
Mehwish Alam 1: Prompting Language Models for Linguistic Structure (Firas GABETNI)
Mehwish Alam 2: Time-Aware Language Models as Temporal Knowledge Bases (Katia Chardon)

2023-10-31

No class

2023-11-07

Pierre-Henri Paris 1: Fine-Grained Analysis of Propaganda in News Articles (Priscille Erulin)
Goran Frehse 2: Safe Model-based Reinforcement Learning with Stability Guarantees (Rémi Gastaud )

2023-11-14: Talks by lecturers

Fabian Suchanek (Télécom Paris): A hitchhiker's guide to ontology

Abstract

Language Models have brought major breakthroughs in natural language processing. Notwithstanding this success, I will show that certain applications still need symbolic representations. I will then show how different methods (language models and others) can be harnessed to build such symbolic representations. I will also introduce our main project in this direction, the YAGO knowledge base. I will then talk about the incompleteness of knowledge bases. We have developed several techniques to estimate how much data is missing in a knowledge base, as well as rule mining methods to derive that data. I will then present our work on efficient querying of knowledge bases. Finally, I will talk about applications of knowledge bases in the domain of speech analysis and the digital humanities, as well as about our methods for explainable AI.

Philippe Xu (ENSTA): Connecting text and images

Abstract

In this talk, we will go through the evolution of computer vision techniques and see how recent breakthroughs in natural language processing are impacting this field of research.

Xi Wang (École Polytechnique): Towards Computational Cinematography from Aspects of Computer Vision, Computer Graphics and Robotics

Abstract

In this talk, Xi WANG's research work at the intersection of robotics, 3D vision, and computer graphics will be presented. His focus is on developing novel techniques in 3D computer vision, virtual camera control, cinematography and even generative models to create innovative multidisciplinary areas. These areas include computational cinematography, neural network-based camera control and tracking, and content generation by leveraging 3D, camera and other modalities of information. Xi WANG holds a Ph.D. that was supervised by Marc Christie and Eric Marchand at Univ Rennes, IRISA, Inria Rennes. He is currently a postdoctoral researcher at LIX, École Polytechnique, under the guidance of Vicky Kalogeiton.