Annonce en lien avec l’Action/le Réseau : aucun
Laboratoire/Entreprise : Gipsa-lab
Durée : 2 years
Contact : simon.barthelme@gipsa-lab.grenoble-inp.fr
Date limite de publication : 2018-10-31
Contexte :
Graph Signal Processing (GSP) is a nascent field that aims to generalise tools used to process time series (line graphs) or images (regular 2D grids) to the setting of more general graphs. Examples include sensor networks, road networks, social networks, etc. One of the central ideas in GSP is to define an equivalent of the Fourier transform via the graph Laplacian, which opens up many ways to filter out noise, estimate signals, etc. From a statistical viewpoint, GSP techniques take graph structure as given, which is suboptimal. Indeed, in many cases graphs are imperfectly observed: missing or spurious links, missing nodes, and sometimes even entire parts of the graph that are missing. How then should we interpret the graph Laplacian and the methods based on it?
Sujet :
The goal of the post-doc is to explore the robustness and interpretability of GSP techniques under noisy structures. We will focus especially on the links between GSP and a special case of Gaussian processes called Gauss-Markov Processes.
The candidate will work with Simon Barthelmé and Ronald Phlypo, and be based in the Signal and Image Processing department of Gipsa-lab (VIBS team).
Profil du candidat :
We are looking for candidates with a background in statistics, signal processing, or machine learning. We will also consider candidates from statistical physics.
Formation et compétences requises :
The candidate must hold a PhD or be very close to completing one.
Please send a CV, a motivation letter, and at least one recommandation letter to simon.barthelme@gipsa-lab.fr and ronald.phlypo@gipsa-lab.fr. If you are about to finish your PhD, please supply a certificate by your supervisor or your doctoral school that mentions the date of your prospective defense.
Adresse d’emploi :
Gipsa-lab, Grenoble, France