The position is associated to the MUCKE project, funded within the CHIST-ERA funding scheme of the FP7, which fosters highly innovative and multidisciplinary collaborative projects in information and communication sciences and technologies. The project started in October 2012. Its main objective is to devise new and reliable knowledge extraction models designed for multilingual and multimodal data shared on social networks.
The post-doctoral researcher will first work on multimodal concept representation (around 2/3 of the post-doc duration) and then on user credibility estimation for multimedia flows (around 1/3 of the post-doc duration). The first task will focus on the proposition of innovative language and image processing techniques as well as on their combination using information fusion techniques. Particular attention will be given to the scalability of the methods developed because one important project objective is to model a large number of concepts (hundreds of thousands) and an important volume of textual and visual documents (tens of millions). For images, the main problems tackled in MUCKE are parsimonious local image description, large scale concept detection and concept visualness. Multimedia fusion will be based on probabilistic models that move beyond a scoring of similarity towards providing an actual probability of relevance. For textual documents, the objective is to depart from text representations as bags of words (unigrams) to representations as vectors of unambiguous concepts and entities through concept and entity recognition and disambiguation.
The second task will focus on the proposition of a novel model for estimating user credibility for multimedia flows, which combines content and structural features of the user's networked representation. Using results obtained for concept representation, credibility estimation will be cast as an information fusion problem, with a first step that combines textual and visual information and a second step that combines user content and context. Another important credibility related question that will be investigated is whether credibility should be a global characteristic of a user or domain related.
Candidates must hold a PhD in Computer Science obtained less than two years ago or expect to complete it before December 2013 and must be under 30 years old. Experience in Natural Language Processing, Information Retrieval or Computer Vision is expected, as well as willingness to learn new topics related to the post-doc.
The lab is situated in Palaiseau, 20 km south of Paris and is part of the CEA LIST Institute, which focuses its research activities on developing innovative technologies for smart and complex systems. The post-doc will be part of the lab's multimedia team, which consists of researchers in NLP, computer vision and multimedia processing. The working language is English. A more detailed description of LVIC's activities is available at http://www.kalisteo.fr/en/index.htm.
For further information, please send an expression of interest to adrian.popescu@cea.fr, attaching a CV and a cover letter. The position is available from October 2013 and is open until filled.
Adrian Popescu
CEA LIST
Vision and Content Engineering Lab
Centre de Saclay - NanoInnov Bât. 861
adrian.popescu@cea.fr
http://comupedia.org/adrian
No comments:
Post a Comment