SuperMann Meets Proximal Envelopes: From Embedded to Big Data Optimization

The exponential increase in computing power as predicted by Moore's law in combination with the need for improved performance in engineering systems as well as the ever-increasing volume of stored data have rendered optimization ubiquitous in almost every branch of engineering. From embedded control and signal processing to data mining and machine learning, optimization problems arise everywhere. However, the challenges that need to be overcome vary with respect to the specific domain. The key to success of embedded control techniques such as model predictive control, lies in the ability to accurately solve small scale problems within micro/nano-seconds with low computational power. In big data applications the needs are shifted towards scalability of algorithms for achieving the specific task in reasonable time. These conflicting objectives are in stark contrast with the overarching goal of making optimization tools available to non-experts; choosing the right algorithm seems to be as hard as solving the problem itself. The main goal of this project is to develop a unifying algorithmic framework that can achieve the specific goals of the corresponding application with minimal effort by the user. This will be made possible due to ongoing theoretical advancements achieved by the research team of the promotor: novel interpretations of classical methods pave the way in developing very efficient optimization algorithms for the full range of the aforementioned applications.