[Corpora-List] Postdoc at CEA and LIMSI on multimedia semantic disambiguation

Brigitte Grau brigitte.grau at limsi.fr
Thu Mar 22 10:24:38 CET 2018


*CEA LIST and LIMSI, CNRS propose a post-doc on multimedia semantic disambiguation. The subject is related to computer vision and natural language processing, and aim at creating common spaces for representing both text and image.*

*12-month post-doc at Paris-Saclay, to be filled in June 2018.*

*Job Description*

Entity linking consists in linking mentions of entities of a text to entities of a knowledge base if they exist, in order to normalize these mentions. It can be useful, for example, for populating knowledge bases by extracting information from texts. One important issue relates on the resolution of ambiguities because the systems have to choose from a large number of candidates.

The disambiguation of entities currently concerns only textual data, and relies on NLP and knowledge representation methods. The proposed subject aims to study the interest of adding the visual dimension to help disambiguation, whenever it can be useful.

Computer vision has experienced a significant break due to deep learning architectures. Deep learning has also had a beneficial impact in the field of natural language processing. In particular, word embedding models offer a representation of the textual modality that allowed significant advances in several tasks. This topic is thus at the intersection of computer vision and natural language processing. The proposed approach is based on an early coupling of visual and textual entities, obtained within a common image-text space resulting, for example, bya neural network that learns to optimizethe alignmentof the documents to be linked (image and corresponding text). The learned embeddings will be usedto improve the selection of candidate entities from a mention in a text. In order to evaluate the method developed, wewill tacklea task of entity linkingwithin illustrated press articles. Given the novelty of the problem, existing corpora may be enrichedthen releasedto the scientific community. *Candidate Profile*

Thecandidatesshould possess a doctoral degree in computer vision, machine learning or natural language processing, together with a solid mathematical background, good programming skills and a strong academic publication record. The starting date is June2018. The contract is for a 1-year fixed-term period with a possibility for 1 yearextension at CEA Paris-Saclay. The salary is 37-41k€ depending on the diploma and experience.

Please apply by sending by mail a CV, a focused motivation letter and one to three recommendation letters to _herve.le-borgne at cea.fr <mailto:herve.le-borgne at cea.fr>_and _bg at limsi.fr <mailto:bg at limsi.fr>_.

---

Laboratory website : _http://www.kalisteo.eu/en/index.htm __and https://www.limsi.fr_

HLB : _https://sites.google.com/site/hleborgne/_

BG : _https://perso.limsi.fr/bg/PageBrigitteGrau-en.html_

-- http://perso.limsi.fr/Individu/bg/ Groupe ILES - LIMSI Bât. 508, rue John von Neumann 91405 ORSAY Cedex tel. 01 69 15 80 03

et

ENSIIE Ecole Nationale Supérieure d'Informatique pour l'Industrie et l'Entreprise 1 square de la résistance, 91025 EVRY Cedex tel. 01 69 36 73 44

-------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: text/html Size: 7751 bytes Desc: not available URL: <https://mailman.uib.no/public/corpora/attachments/20180322/77052bb6/attachment.txt>



More information about the Corpora mailing list