This position aims at relating modern insights in (Online) Machine Learning (OML) to the esteemed tool of the Kalman Filter (KF). OML has led to renewed insights in such tasks as optimization, multiple-armed bandit algorithms and sequential designs, and comes nowadays with a solid theoretical underpinning. The KF on the other hand is the workhorse of the control-engineering, and is for example used as the observer of choice in many state-feedback control schemes.
Integrating KFs in a modern OML setting would provide new insights and suggest new extensions to basic schemes. The particular focus would be how to extend the KF to the case of many variables. Modern techniques of OML have excelled in this topic, while traditional approaches as the KF have not been studied extensively in this regard.
This project requires a more mathematical oriented background. That is, we invite people with a background in mathematical physics, control engineering or statistical sciences.
Please see further description, and apply via: