Skip to content

RLposition.md

Latest commit

 

History

History
66 lines (47 loc) · 6.29 KB

RLposition.md

File metadata and controls

66 lines (47 loc) · 6.29 KB

Here are some details about the position. I've tried to avoid jargon and go straight to the point with relevant details. Of course what is written below are my own views and are given only to help candidates quickly get acquainted with the position.

-- Emmanuel Rachelson (Feb 24, 2019)

General stuff about ISAE-SUPAERO

How it works: professors at ISAE-SUPAERO don't have a fixed teaching commitment, as in universities. Our job on the "teaching" side is to insure that the training goes well and stays at the best level possible (this is evaluated through the competitive exams applications and through the alumni's careers afterwards - note that ISAE-SUPAERO is among the 5 first French Grandes Ecoles).

Currently, there are 101 professors at ISAE-SUPAERO, covering all disciplines, for 33 training programs including the flagship Grande Ecole program, and around 1700 students. Practically, it implies that we spend quite a bit of time managing teams of external instructors (almost 2000/year in total) and the associated network, delegating the construction of classes and the training, and coordinating the whole curriculum.

Consequently, a standard teaching volume for an ISAE-SUPAERO prof is around 60-80 hours/year with a high variance across professors (depending on their age, discipline, personnal preferences, etc.).

Concerning research, ISAE-SUPAERO hosts research in many different disciplines, including AI. SuReLI is the place where most of ML / AI related things happen but many great colleagues contribute in their respective application fields too. Professors have (quite reasonable) individual objectives set each year but little pressure. Although there is a strong encouragement for research excellence (and a good environment provided by our institution), our career and salary progression is not as strictly conditionned by our short-term research achievements than in universities.

On the AI side, SuReLI is one of the players in Toulouse with great connections at IRIT, LAAS, IRT Saint Exupéry, with most big private players (Airbus, Thalès, etc.) or within ANITI (the major AI cluster in Toulouse). We are also among the founders of the "Toulouse Interdisciplinary Deep Learning" group.

Teaching duties of this position in particular

The three first classes below are within the Data Science MS program in the Grande Ecole cursus, which is managed by me, so this is close work with me, as for most tasks in this position. The fourth is slightly less important (and also within my perimeter but outside the MS training). I make a point of having happy colleagues and I expect the new professor to be an autonomous contributor to AI related topics (both for training and research), so even though all these responsibilities are somehow related to what I have built over the last years, there should be a healthy balance between common work and autonomy.

Here are the classes that the new professor should manage and take the leadership on:

  • Tools of Big Data. 40h class covering databases (SQL and NoSQL), functional programming, Spark, GPGPU, cloud computing. The class already exists with external trainers and a reasonable scenario would include giving the databases and/or functionnal programming parts.
  • Digital Economy and Data Uses. 30h. Business models, privacy and security, dataviz. The class exists with external trainers and there is no need to give it. Possible evolutions are given below.
  • Hackathon. We run a 3-days data science hackathon each year, with external companies as partners.
  • Introduction to Big Data. 20h. Introductory class. It's mainly a task of managing the training team.

Of course, once the classes are under the responsibility of the new prof, he/she has the duty of keeping them at the best level possible, which might involve changing the contents, etc. But we generally do that together.

Here are extra hours that can be shared with the new professor if he/she wishes (no obligation here):

  • Algorithms in Machine Learning. 80h in total. My big class that goes from statistical learning to hands-on algorithms. Of course it covers Deep Learning and we have 20h on Reinforcement Learning. I'll be happy to delegate or share some classes (if not full parts of the program).

Finally, there are some student projects that should be tutored, within the Grande Ecole cursus and within the AI & Business Transformation executive Master that should start next september.

Here are the projects for the future that the new professor should be part of:

  • Evolution of the Data Science cursus. Basically my current plans involve simplifying things inside the cursus and reshaping some modules to better fit our training goals.
  • Advising professors in other disciplines to help introduce AI content to their training programs (only advising, no teaching involved here).
  • Participate in the evolution of the AI & Business Transformation executive Master.

Research environment

Here are a couple of current hot topics in SuReLI that form a good basis for understanding what we do and where we want to go:

  • Coupling model-based optimal control and (Deep) RL in robotics,
  • Modular (Deep) RL for AGI and tranfer between tasks.
  • Discovering entities and objects to structure learning in RL and allow transfer.
  • Life-long and continuing transfer in Deep RL / avoiding catastrophic forgetting.
  • Fundamental questions in RL: off-policy learning, sample efficiency, bias-variance tradeoff (and others as they appear).
  • Model uncertainty robustness in RL and transfer between models.
  • Parameter control in Optimization processes via RL

Other profs / researchers close to SuReLI:

  • Carlos Aguilar (ISAE-SUPAERO, RL, cryptology)
  • Nicolas Mansard (LAAS CNRS, robotics)
  • Daniel Delahaye (ENAC, optimization, evolutionary algorithms)
  • Olivier Sigaud (ISIR / Paris Sorbonne, RL, robotics)
  • Nicolas Perrin (ISIR, RL, robotics)
  • Sylvain Cussat-Blanc (IRIT, evolutionary algorithms, artificial life)
  • Hervé Luga (IRIT, evolutionary algorithms, artificial life)
  • Rufin VanRuellen (Cerco CNRS, deep learning and connection with the human brain)

A little further but still related:

  • Yves Brière (ISAE-SUPAERO, Control Theory)
  • Caroline Chanel (ISAE-SUPAERO, Planning under uncertainty)
  • Frédéric Dehais (ISAE-SUPAERO, Neuro-ergonomy)
  • Michael Bauerheim (ISAE-SUPAERO, Fluid mechanics)