PostDoc on road traffic datamining at Mines ParisTech

PostDoc on road traffic datamining at Mines ParisTech

Posted by Rebecca Martin on Fri, 16/01/2009 - 12:13

Applications are invited for a post-doctoral position in road traffic datamining and prediction, for 15 month starting within S1 2009, at Robotics Lab. of Mines ParisTech (Paris, France).

The robotics laboratory (CAOR) of Mines ParisTech, associated with IMARA project of INRIA in LaRA « Joint Research Unit », has been involved in 2 big European projects (REACT and COM2REACT) using V2I (Vehicle toInfrastructure) and V2V (Vehicle To Vehicle) communications for enhancing global « road information system ». In these projects the 2 labs worked in particular on analysis and prediction of traffic for improving preventive re-routing strategies in order to reduce congestions. An algorithm has been developed for reconstructing and predicting traffic from a fleet of « sensor » vehicles regularly sending position/speed/traffic information.
CAOR has just started, again with IMARA, and together with TAO project of INRIA and LET of Lyon, a new collaborative project sponsored by ANR (French national research funding agency). This project will focus on analysis and prediction of road traffic, first on realistic simulated data (to be produced with Metropolis software developed by LET), then on real data.

Research work description
The work to do is firstly data-mining of traffic data, seen as a graph whose each edge is a road section with associated traffic level (mean speed, travel time or congestion level), in order to extract common traffic patterns. For this « pattern mining », the idea is to test various clustering methods, in particular unsupervised training algorithms, such as Kohonen maps and K-means, so as to identify main « attractor » states and/or usual traffic states.
Then, the candidate should try to build a simplified dynamic model, as prediction of transitions between identified patterns. In particular, the possibility to exploit fully the graph structure of roads network shall be examined, by experimenting “graph kernel methods” recently developed and mainly applied in the context of bioinformatics.
A possible extension is analysis of road network as a complex dynamical system (bifurcation diagram, etc…).

The candidate should hold a good PhD in the field of statistical machine-learning and/or data-mining, with:
• Very good knowledge of data mining and analysis techniques, as well as of machine-learning methods;
• Good knowledge in probabilities and statistics (in particular Markovian models);
• Some knowledge on graphs and associated algorithms;
• Good computer programming skills (C/C++/Java)

Speaking French is not absolutely mandatory, but would be a plus.

Duration and date
Duration of post-doctoral contract is 15 month, starting within first semester 2009.

Supervision and contact :
Fabien Moutarde, (+33), Fabien.Moutarde (at)

To apply:
Candidates must send a detailed CV, with a cover letter, main publications (or links), together with name and contact of at least 2 references, to above e-mail address.