This page (like this whole web site) is now (January 2023) totally obsolete.
Please check the new "Job Offers" page of the TAU team


Currently, there are no open PhD positions, please come back later - though we usually have more open applications than available positions.
Thank you for your interest in TAU

• Machine Learning and Optimization for Long Term Investment Planning
  contact Olivier Teytaud (INRIA-LRI)
  More info: http://www.lri.fr/~teytaud/metis.html
  Optimization; Energy; Machine Learning; MDP solving

---

• Collaborative development in planing
  Funded by the European project MASH
  More details here: http://teytaud.over-blog.com/article-36285588.html

---

• Monte-Carlo Tree Search (for people ok for travelling)
  Funded by a ANR project IOMCA
  includes travels in Taiwan (lodging + plane funded)
  More detailes here http://teytaud.over-blog.com/article-35978426.html

---

• Robust Stochastic Continuous Optimization (September 2009)
  Funded by a ANR project OMD2
  Details will be posted soon - contact Anne Auger

---

• Multi-objective optimisation of automobile engine (was October 2008)
  Funded by a CIFRE scholarship (PSA Peugeot Citroën)
  [+]
  • The context of the PhD is the optimisation of the combustion chamber in an automobile engine. To day, a first approach tackled the single objective problem, by coupling a 3D combustion simulator with a single objective Genetic Algorithm. The restults were industrially relevant, and have been obtained within at a reasonnable CPU cost.
    However, most combustion problems involve in fact several contradictory objectives (comsumption, driving comfort. pollution (NOX, dust, ...). Most existing multi-objective algorithms will require too many calls to the objective functions, and will hence be too CPU intensive to be of practical use in this context.
    The goal of the PhD is to design multi-objective optimization algorithms that will require only a limited amount of calls to the objective functions, either by using/adapting existing techniques, such as surrogate models, or by imagining new algorithms, using for instance recent techniques of Machine Learning.
  • Send applications to marc [dot] schoenauer [at] inria [dot] fr

---

• Learning in collective robotics (was January 2008)
  see details

---

• Automatic tuning of search and optimization algorithms: (was July 2007).
  see details

---

• Approches évolutionnaires et hybrides pour la replanification d'horaires de trains: (was May 2007).
  Advisor Marc Schoenauer and Christelle Lerin (SNCF)
  Funded by a CIFRE scholarship (SNCF)
  [+]
  • The local reconstruction of a railway schedule following a small perturbation of the traffic, seeking minimization of the total accumulated delay, is a very difficult and tightly constrained combinatorial problem. Notoriously enough, the railway company public image degrades proportionally to the amount of daily delays, and the same goes for its profit!
    Evolutionary Algorithms using an indirect representation based on a semi-greedy scheduler that incorporates the expert domain knowledge have been demonstrated to perform from much worse to much better than the commercial tool CPlex), depending on the incidents.
    A first research direction is to try to identify the different types of incident and to a priori detect which tool is better suited than the other
    The second direction is that of hybridization of both thechniques, as premilinary experiments have shown that on the instances that CPlex fails to solve rapidly, initial approximation by an evolutionary algorithm allows CPlex to then rapidly find the optimal solution.
  • Strong background in optimization is mandatory. Knowledge of MIP problems and/or evolutionary algorithms will be appreciated. Good programming efficiency is essential.
  • Send applications to marc [dot] schoenauer [at] inria [dot] fr

---

• Segmentation et évolution artificielle pour la planification temporelle: (was Dec. 2006).
  Advisor Marc Schoenauer and Pierre Savéant (Thalès)
  Funded by a CIFRE scholarship (Thalès)
  [http://www.lri.fr/~marc/TheseDE.html|+]

---

• Learning Surrogate models: (was March 2007).
  Funded by INRIA (CORDI). Details and application

---

• Evolutionary Generation of Test Data: (was Nov. 2006).
  Advisor Marc Schoenauer
  Funded by the European FP6 project EvoTest (Automatic Generation of Test Data for Complex Systems, http://complexsystems.lri.fr/evotest).
  [+]
  • The first work of the PhD student will be to design a tool for the automatic generation of an Evolutionary Algorithm from test data structure, by extending an existing graphical prototype. The research area is that of representation-independent variation operators.
    The second part of the work deals with the automatic tuning of EA parameters. Research will concentrate on both off-line algorithms (including meta-EAs, ANOVA experiments, learning accross runs), and on-line algorithms (racing algorithms, self-adaptation, Sequential Parameter Optimizatin, entropy-based mechanisms, ...).
    Knowledge of Evolutionary Algorithms is of course highly desirable. Experience and skill in Java and C++ programming are mandatory, as the project requires imlementation of prototypes as well as interface with industrial software frameworks. Additional knowledge in Machine Learning and/or Program Testing will be appreciated.
    Send applications to marc dot schoenauer at inria dot fr

---

• Active Regression: (was Oct. 2006).
  Advisor Michèle Sebag
  Bourse Digiteo

---

• Machine Learning for Evolutionary Robotics : was April 2006
  Advisors Michèle Sebag and Marc Schoenauer
  A CORDI position (i.e. available in priority to students from abroad).

---

• Machine Learning and EGEE: was April 2006
  Advisors Cécile Germain and Michèle Sebag
  A CORDI position (i.e. available in priority to students from abroad).