Coordination Funds

点击翻译按钮获取中文摘要

Within this research unit, we will study the algorithmic challenges of fundamental artificial intelligence problems in two central areas of geodesy that both deal with geometric abstractions of the real world: cartography and physical geodesy. A central characteristic of our approach is a focus on an inherently geometric data representation motivated by the aim to study the problems in their native form. The systematic study of the underlying algorithmic challenges of AI problems is a central task in the field of algorithmic data analytics. This research unit thus brings together experts in the area of algorithmic data analytics and in the above two mentioned areas of geodesy. As a mediator between the two ends we see the fields of computational geometry and algorithm engineering, in particular the way to approach problems that these fields have established. Computational geometry transfers geometric real-world data into discrete configurations, thereby enabling the application of discrete optimization techniques. Algorithm engineering brings a whole routine of transferring knowledge between theoretical analysis and practical applications. As project outcome we expect to obtain new data analytics methods that deal with the specific challenges of geodetic AI problems. Our new methods will be tailored to the given problems and usually be accompanied with provable performance guarantees. They may use new algorithm engineering and combinatorial optimization techniques building on the geometric structure of the input. An ongoing experimental evaluation on practical real-world data from geodesy will steadily lead to theoretical as well as practical improvement of the applied methods. Our central goal is to substantially close the gap between current research in artificial intelligence and geodesy and to establish persistent links between the two disciplines. This will allow future research on automation in geodesy to be conducted on a more solid algorithmic basis. At the same time, our research unit will advance the algorithmic understanding of machine learning tools for geometrically represented data and corresponding distance measures. We will develop a collection of carefully designed algorithms and implementations that are able to exploit the geometric structure of the data even for very large data sets, under geometric side constraints and for multiple objectives. We are convinced that our approach will not only be interesting to researchers in geodesy but to any other area that involves the analysis of geometric data.