Sie sind hier: Startseite Kolloquium Towards Automated AI via Data Stream Mining and Automated Algorithm Selection

Towards Automated AI via Data Stream Mining and Automated Algorithm Selection

Prof. Dr. Heike Trautmann, Depart. of Information Systems, University of Münster

Art des Termins
    Wann 21.06.2022
    von 11:30 bis 12:30
    Wo Room no. 0.016
    Termin übernehmen vCal



    Forschungsvortrag: Towards Automated AI via Data Stream Mining and Automated Algorithm Selection

    After an overview of my core research areas in the Data Science and AI domain, specifically data stream mining and online media analytics, this talk will focus on Automated Algorithm Selection and its crucial importance for constructing high-performing automated AI systems. It has long been observed that for practically any computational problem that has been intensely studied, different problem instances are best solved using different algorithms. This is particularly pronounced for computationally hard problems, where in most cases, no single algorithm defines the state of the art; instead, there is a set of algorithms with complementary strengths. This performance complementarity can be exploited in various ways, one of which is based on the idea of selecting, from a set of given algorithms, for each problem instance to be solved the one expected to perform best. The task of automatically selecting an algorithm from a given set is known as the per-instance algorithm selection problem and has been intensely studied over the past 15 years, leading to major improvements in the state of the art in single-objective continuous black-box optimisation and in solving a growing number of discrete combinatorial optimisation tasks, including traveling salesperson problems.



    Lehrprobe: Data Analytics (Unsupervised Learning) — Introduction to Data Stream Clustering

    Within a master course on Data Analytics with focus on unsupervised learning, I will give an introduction to data stream clustering as an important technique in data stream mining for e.g. customer segmentation, analysing sensor data or topic modeling in social media. Our modern world produces an endless stream of data while the sheer volume and speed has surpassed our ability to process, analyse, store and understand it properly, especially when requiring real-time systems. A conceptual introduction of the Data stream model will be followed by a taxonomy of stream clustering approaches and a motivating example on customer segmentation on transactional data.


    Location: Institute of Computer Science, Friedrich-Hirzebruch-Allee 8, 53115 Bonn, Room no. 0.016