Regularisation via Von Neumann Entropy for Matrix Completion

点击翻译按钮获取中文摘要

Matrix completion is an area of machine learning with very immediate and practical applications in a business setting. The most notable being recommendation systems such as the so-called 'Netflix problem' in which a user's preferences for a given set of objects are inferred under the assumption that the data may be represented by a partially-known low-rank matrix. The aim of this research is to develop new, efficient algorithms with low mistake bounds for matrix completion in the context of online learning with a focus on entropic regularisation techniques. There's a possibility that parts of this research may be done in collaboration with Siemens UK. This research aligns very closely with EPSRC's research area of Artificial Intelligence Technologies.