Today, sensors are widely used in different environments (cars, windmills, factories, etc.) and emit data very often, e.g., every second. This results in millions of very large time‐series. Such time‐series are not well‐supported by the current Big Data tools. For example, there is a need for advanced statistical methods that provide more insights than the traditional drill‐downs and roll‐ups. The user would also benefit greatly if historical and current data could be used to predict the future readings. The vision is thus to create a BI system that enables efficient storage and analysis of time‐series. This system will allow seamless unified querying of historical, streamed, and predicted data by using models. At the logical level, the data will be presented in a multidimensional view, but at the physical level the models will be represented in non‐relational forms. Further, the developed algorithms will consider the trade‐off between query accuracy and response time such that the user can choose to wait less for a less precise result or wait longer for a precise result. This topic thus involves novel solutions both for the storage layer and for algorithms for query processing.

Main Advisor at Aalborg Universitet (AAU)
Co‐advisor at Technische Universität Dresden (TUD)