Ce projet de recherche doctorale est publié a été réalisé par Patrick Gallinari

# Description d'un projet de recherche doctoral

### Learning representations in structured domains using neural networks and applications to social network analysis

**Mots clés : **

**Directeur de thèse** :
Patrick Gallinari

**Unité de recherche** :
Laboratoire d'informatique de Paris 6

**Ecole doctorale** :
École Doctorale Informatique, Télécommunications, Électronique de Paris

**Domaine scientifique principal**: Divers

### Résumé du projet de recherche (Langue 1)

### Résumé du projet de recherche (Langue 2)

-* Learning representations for heterogeneous networks and multimodal data

Develop a representation learning framework adapted to complex relational data, representative of what can be found on heterogeneous social networks. This framework will exploit earlier developments on manifold learning [Weston 2008]. Both non parametric and parametric method will be developed with a focus on the latter which are more suitable for performing inference in evolving environments.

-* Deep Neural Networks

An alternative direction is the use of Deep Neural Networks which can be used to implement parametric functions for learning to encode data onto latent representations. Deep Neural Networks allow us to push further the learning of representation by building non linear transformations through the combination of successive mappings learned sequentially.

-* Integration of temporal and dynamic information

Social data are dynamically evolving. This dynamic or temporal evolution is a major characteristic of these data which should be considered in most applications. The above models have been mainly developed for static unstructured or structured data and the integration of their dynamic nature into the above models is another direction that will be considered in this proposal.

-* Applications in the domain of social network analysis

These methods will be applied in different applications for the analysis of social data. We forecast two main applicative domains which are respectively social recommendation (i.e. the development of social recommendation systems that take into consideration the social links between the users) and classification and ranking in social networks which is more of a generic problem that can be used for different applications.

### Informations complémentaires (Langue 2)

-* [IBM 2013] "IBM What is big data? — Bringing big data to the enterprise". 01.ibm.com

-* [ICLR 2013] International Conference on Learning Representations 2013, https://sites.google.com/site/representationlearning2013/

-* [Socher 2011] [Socher R., Lin C., Ng A. Y., Manning C.D., Parsing Natural Scenes and Natural Language with Recursive Neural Networks, ICML 2011.

-* [Le 2012] Le Q. V. , Ranzato M., Monga R., Devin M., Corrado G., Chen K., Dean J., Ng A., Building high-level features using large scale unsupervised learning, ICML 2012

-* [Lee 1999] Lee D.D., Seung H.S. "Learning the parts of objects by non-negative matrix factorization". Nature 401 (6755), 1999.

-* [Poultney 2006] Poultney C. , Chopra S. , Lecun Y., Efficient learning of sparse representations with an energy-based model, NIPS 2006

-* [Bordes 2011] Bordes A., Weston J., Collobert R., and Bengio Y.. Learning structured embeddings of knowledge bases. In AAAI, 2011.