For DAS team
Since September 1, Dr. Florian Sobieczky reinforced the Data Analysis Systems team. He studied mathematics and physics. He focuses his research on the topic Predictive Modeling.
Dr. Florian Sobieczky delivers insight in his research topics
As a researcher in data-science, I am currently mostly interested in the statistics of data assessment problems, as they occur in typical real-world predictive modeling tasks in manufacturing environments. For example, processing time-series of independent sensors into a data table allowing classification and regression often requires balancing the sequences of measured values in a meaningful, non-intervening way. It is my goal to develop a theoretical framework for and implementations of preprocessing tools for large incoherent data sets.
I am also interested in structural equation modeling (SEM) of the manufacturing processes, especially for production lines. Finding the causal relationships of mechanisms corresponding to assessed data can be difficult, even if the order in time of these production steps is known. For example, it may be difficult to decide from the assessed data which of two serially arranged production machines is responsible for an observed increase in wastage. The phenomenon (called Mediation) is due to the possibility of indirect effects, such as the hidden triggering of malfunction of the hindmost machine by the front-most.
My background is that of math and physics. Having worked in the field of discrete probability, I like searching for solutions for the above mentioned tasks by using models from percolation, queueing theory, and random walks on graphs. For example, the variability in performance of various predictive models under perturbations of the incoming data can sometimes be compared to that of a perturbed random walk on a graph. I am interested in using such analogies to predict the effects of unbalanced data, perturbed measurements, or ill-configured production machinery.