Sparsity and structures in large-scale machine learning problems

This project aims at investigating, both from a theoretical and computational point of view, the design of novel model-based and data-driven feature extraction and sparsification strategies for the approximation of complex Machine Learning problems. Special attention will be drawn to kernel-based and artificial-neural-network-based methods, which are two of the most important families of modern Machine Learning algorithms.

The terminology "big data" is generally used to refer to datasets that are too large or complex for traditional data-processing technics to adequately deal with them. The exploration of such datasets with modern Machine Learning techniques therefore raises many theoretical and numerical challenges. The numerical complexity inherent to the processing of datasets indeed generally grows polynomially with their size, compromising de facto the analysis of very large datasets. In addition, the treatment of complex datasets often results in models involving a large number of parameters, making such models difficult to train while limiting their interpretability and increasing the risk of over/underfitting. Since such large-scale and complex datasets are more and more common in nowadays big-data and real-time-analytic era, their efficient processing is of great importance, not only at from purely scientific point of view, but also for many industrial and real-life applications.