Duration: 3 years
Category: Research
Type: PhD thesis
Location: Institute for Infocomm Research - A*STAR - Singapore Laboratoire ETIS UMR 8051 CNRS/CYU/ENSEA

The MIDI (Multimedia Indexing & Data Integration) Team, ETIS UMR 8051 Laboratory, CY Cergy Paris University, ENSEA, CNRS (France) and the Institute for Infocomm Research, A*STAR (Singapore) launch a call for applications for a doctoral position in “Representation learning for multimodal data”.

The context

The PhD student will be involved in an international collaboration research between CY Cergy Paris University (France) and Institute for Infocom Research A*STAR (Singapore) spending 18 research months at ETIS Laboratory in France and 18 research months at A*STAR in Singapore.

Beginning of the Thesis: As soon as possible

Project description

The performance of machine learning algorithms is largely determined by data representation, which is due to the fact that different representations may entangle and hide distinct explanatory aspects of variation behind the data to varying degrees. The problem using these types of representation learning on heterogeneous multimodal data, including time series, text, images, etc. present methodological issues in the modelling and learning from such complex data. This challenge renders the supervision process almost impracticable and very uncertain given the unknown dynamical nature of the observed systems.

We want through this thesis to explore the unsupervised topological learning of multimodal data presenting a complex structure allowing to learn their representations. We are particularly interested in heterogeneous data whose representation may have been informed in different ways: expert representation which may be complex (for example: dynamic multi-graphs which may possibly have different topologies for each observation of the dataset and for every moment) or automatically learned representation (for example: embedding in a high-dimensional space with dense vectors and potential correlations between the components), images, signals.

The topological learning models based on neural networks and probabilistic models will be analyzed in this project and the expected results are proposal of representation learning approaches able to learn the topological information from multi-modal data by increasing the reconstruction error. 

Co-directors of the Thesis:

Nistor Grozavu, Full Professor
ETIS UMR 8051, CY Cergy Paris University, ENSEA, CNRS, France
Web : http://www.grozavu.fr

Xiaoli LI, Full Professor 
Institute for Infocomm Research, A*STAR, Singapore.
Web: https://personal.ntu.edu.sg/xlli/

Shared ARAP scheme funding CYU (France) – A*STAR (Singapore)

  • Salary: 18 months CYU contract (legal doctoral salary, around 1500-1600€/month) + 18 months A*STAR contract (S$ 2700/month)
  • Running cost: 1000 €/year
  • Round trip to the other country: 600 – 1000 €
  • Per Diem in the other country: 100€
  • Intl conferences (at least 3): 2000€ including flight + registration + housing * 3


ETIS UMR 8051, CY Cergy Paris University, ENSEA, CNRS, France


Institute for Infocomm Research, A*STAR, Singapore.

To apply:

Complete application with CV, recommendation(s) letter(s), academic results to be sent to:
email : nistor.grozavu@cyu.fr

The details of the research subject can be downloaded here.

Apply for this position

Allowed Type(s): .pdf, .doc, .docx