Scalable machine learning for big data


Abstract : The increasing need of dealing with big datasets requires a new way of designing, implementing, validating and disseminating machine learning algorithms. The research will tackle these aspects by assessing which aspects of state-of-the-art machine learning can already scale-up and what needs a redesign/rethinking in order to deal with new programming (e.g. Map-Reduce, Spark) and platform (e.g. Hadoop) paradigms. A specific attention will be devoted to: - assessing the limits of transposing conventional machine learning to the map-reduce formalism. - design statistical, computational and programming solutions in order to make machine learning algorithms usable in distributed big data settings - define performance models for comparing learning solution on conventional and big data platforms - propose new programming and computational formalisms for effective machine learning in big data - propose solution to the problem of making big data machine learning algorithm and benchmarks accessible to the largest scientific community (e.g. in bioinformatics). Note that no specific funding is available for this proposal. Please contact me only if you have already a funding source.
Promoteur/Supervisor : Prof. Bontempi Gianluca
Email : gbonte@ulb.ac.be
Site Web/Web site : mlg.ulb.ac.be
Centre de recherche/Research center : Machine Learning Group and (IB)^2
Faculté/Faculty : Faculty of Sciences/Faculté des Sciences
Ecole doctorale/Graduate Colleges : Science/Sciences
Ecole doctorale thématique/Graduate School (French Only): Computational Intelligence and Learning (CIL)



Retour à la liste principale/Back to main list