Research Team: MAGNET
Title of the proposal: Learning from Graph Structure and Node Content in Large-Scale Random Networks Keywords: Machine learning, random networks, probabilistic graphical models, spectral methods, graph mining Scientific Supervision: Antonino Freno, Marc Tommasi About Inria and the job: Inria, with its academic, institutional and industrial partners, is commited to major research and innovation projects in the field of computational sciences. The institute disseminates across France thanks to its eight research centres. The Inria Lille - Nord Europe research centre, inaugurated in 2008, employs 357 people, including 250 scientists, in its seventeen research teams. Recognised for its significant contribution to the social and economic development of the Nord - Pas-de-Calais region, the Inria Lille - Nord Europe research centre promotes a policy of close cooperation with major businesses and small enterprises. By encouraging synergies between researchers and industrial partners, Inria contributes to the transfer of skills and expertise in computational technologies and provides access to top-level European and international research in order to support innovation and businesses, particularly in the region of Lille. Whether designing innovative software for business or logistics, modelling living cells or fusion plasma, or developing medical simulators or interfaces to facilitate human-computer interaction, our research opens up new possibilities that can revolutionise common practice and contribute to a better understanding of the natural phenomena which surround us. The Magnet project aims to design new machine learning based methods geared towards mining information networks. Information networks are large collections of interconnected data and documents, such as citation networks and blog networks among others. For this, we will define new structured prediction methods based on machine learning algorithms for graphs. Applications include node classification, link prediction, clustering and statistical analysis of graphs in browsing, monitoring and recommender systems, and more broadly large-scale network mining. Target domains cover social and communication networks, e-commerce, and bioinformatics. For more information, you can visit the project homepage at http://team.inria.fr/magnet/. Mission: Due to recent developments of the Web, huge amounts of structured data are becoming increasingly available, most notably in the form of random networks [3]. Online exchange of information often tends to organize itself through some sort of network, where relevant examples include friendship networks (Facebook), customer-product networks (Amazon), co-authorship and citation networks (DBLP, Google Scholar). Massive application of statistical methods is a crucial element of Web technologies such as search engines, spam filters, or recommender systems. But although statistical machine learning has achieved a solid understanding of uncertain and noisy data whenever they can be formalized as fixed-length feature vectors, we are still far from reaching a general consensus on the correct way of modeling networks as statistical objects. As a consequence, a large variety of statistical models have been proposed recently, but none of them has been generally adopted as a standard reference. This project focuses on a novel statistical approach to large-scale network analysis, based on the Fiedler delta statistic. The Fiedler delta statistic has been used to develop models of conditional and joint probability distributions over (large) undirected networks [1, 2]. However, the proposed models only take into account the graphical structure of the network, ignoring instead the information which is often enclosed in the nodes. Such information is typically stored as a vector-space representation of node (text or multimedia) content. The main goal of the project is to extend the approach based on the Fiedler delta statistic so as to account both for graph structure and node content through a unified formalism. In particular, the resulting model will have to support inference not only at the level of network links, but also at the level of node features. In other words, possible uses of the developed model will be not just in link prediction and graph generation, but also prediction of local properties of nodes based on partially specified profiles and connections. Job offer description: While the first part of the project will be devoted to investigating the problem and designing one or more candidate models, a second part will consist in applying the developed techniques to real-world challenges such as social network analysis, modeling user profiles and linking behavior in telecommunication services, and mining protein-protein interaction data. Large-scale data from renowned ICT industry labs will be made available to possibly explore the impact of our results in a real-world setting. Bibliographical references: [1] Antonino Freno, Mikaela Keller, Gemma C. Garriga, and Marc Tommasi: "Spectral Estimation of Conditional Random Graph Models for Large-Scale Network Data". In: Proceedings of the 28th Conference on Uncertainty in Artificial Intelligence (UAI 2012). AUAI Press, 2012, pp. 265–274. [2] Antonino Freno, Mikaela Keller, and Marc Tommasi: "Fiedler Random Fields: A Large-Scale Spectral Approach to Statistical Network Modeling". In: Advances in Neural Information Processing Systems 25 (NIPS 2012). MIT Press, 2012. [3] Mark E.J. Newman: Networks. An Introduction. Oxford University Press, New York (NY), 2010. Skills and profile : Applicants must have already obtained (or be very close to obtaining) a PhD in machine learning, statistics, or related areas. The ideal candidate will have a strong background in probabilistic graphical models, spectral methods, network science, and/or large-scale learning. Good programming skills in at least one language are necessary (ideally Java, but not necessarily). Fluency in spoken and written English is required. Benefits : Duration : 16 months Salary: 2.620,84 Euros gross/month Monthly salary after taxes : around 2.138 Euros (medical insurance included). Possibility of French courses Help for housing Participation for transportation Scientific Resident card and help for husband/wife visa Additional informations: Before applying, please contact the scientific advisor: name.surn...@inria.fr Security and defence procedure : In the interests of protecting its scientific and technological assets, Inria is a restricted-access establishment. Consequently, it follows special regulations for welcoming any person who wishes to work with the institute. The final acceptance of each candidate thus depends on applying this security and defense procedure. For the first selections, please apply before March 2013, 22nd. For more administrative information and the official call, visit http://www.inria.fr/en/institute/recruitment/offers/post-doctoral-research-fellowships/campaign-2013 Antonino Freno _______________________________________________ uai mailing list uai@ENGR.ORST.EDU https://secure.engr.oregonstate.edu/mailman/listinfo/uai
