The Riemannian and non-smooth optimization group (RANSO) gathers researchers from the ICTEAM Institute in Louvain-la-Neuve and the System and Modeling research unit in Liège. The group emerged from a combination of research interests in nonsmooth optimization and optimization on manifolds.
The development of this research area is primarily motivated by specific large-scale applications that have gained much popularity in the recent years in the framework of L1 optimization, stochastic optimization, and statistical learning. Those applications include robust/sparse/structured principal component analysis, statistics on manifolds (e.g. median calculation of positive semi-definite tensors), and low-rank optimization (matrix completion, collaborative filtering, source separation).
The group has developed an expertise in several interrelated algorithmic techniques: random search algorithms, gradient-based nonsmooth optimization algorithms applied to almost-everywhere-differentiable objective functions, smoothing techniques, and optimization techniques over manifolds.
Since 2012, the group has been funded by the Belgian FRFC through grant 2.4585.12 "Nonsmooth optimization on manifolds".
The RANSO group maintains Manopt, a Matlab toolbox for optimization on manifolds.