We are happy to announce our 2 keynote speakers

Thomas Bäck

Algorithms for Simulation-Based Optimization Problems

Many industries use simulation tools for virtual product design, and there is a growing trend towards using simulation in combination with optimization algorithms for tuning simulation input parameters. The requirements for optimization under such circumstances are often very strong, involving many design variables and constraints and a strict limitation of the number of function evaluations to a surprisingly small number (often around one thousand or less).

Tuning optimization algorithms for such challenges has led to very good results obtained by variants of evolution strategies and by variants of Bayesian optimization algorithms. Evolutionary algorithms are nowadays standard solvers for such applications. In the presentation, sample cases from industry are presented, and their challenges are discussed in more detail. Results of an experimental comparison of contemporary evolution strategies on the BBOB test function set for a small number of function evaluations are presented and discussed, and further enhancements of contemporary evolution strategies are outlined.

Our practical examples are motivated by industrial applications. A typical challenge is to find innovative solutions to a design optimization task. Based on a suitable definition of innovative solutions, an application of this concept to an airfoil design optimization task is discussed in the presentation.

Discussing these applications and the variants of evolution strategies applied, the capabilities of these algorithms for optimization cases with a small number of function evaluations are illustrated.

_______________________________________

Frederic Theodor Stahl

Frederic Stahl is Associate Professor in Data Science at the University of Reading and has been working in the field of Data Mining and Knowledge Discovery in Data (KDD) for the last 12 years. In particular he has been working in the research domain of Big Data Analytics. His research interests here are in (i) developing scalable parallel algorithms for building Data Mining models on large volumes of data; (ii) developing algorithms for building self-adaptive Data Mining models for real-time streaming data and (iii) applications in Big Data Analytics. He currently leads a small group of 5 PhD students working on many aspects of Data Mining and Data Stream Mining. In previous appointments Frederic worked as a Lecturer at Bournemouth University and as Senior Research Associate at the University of Portsmouth. He obtained his PhD in 2010 from the University of Portsmouth with the title “Parallel Rule Induction” and his Engineering Diploma in Bioninformatics in 2006 from the University of Applied Science Weihenstephan (Germany). He has published over 50 articles in peer-reviewed conferences and journals. He is heavily involved in the BCS SGAI, the  Specialist Group on Artificial Intelligence of the British Computer Society. Here he serves as an elected committee member, is the main organiser of the UK Symposium on Knowledge Discovery and Data Mining, co-organiser of the societie’s annual International Conference on Artificial Intelligence and Guest Editor of the conference’s Special Issue journal (Expert Systems, Wiley.

Building Adaptive Data Mining Models on Streaming Data
in Real-Time, an Outlook on
Challenges, Approaches and Ongoing Research 

Advances in hardware and software, in the past two decades have enabled the capturing, recording and processing of potentially large and infinite streaming data. As a consequence the field of research in Data Stream Mining has emerged building Data Mining models, workflows and algorithms enabling the efficient and effective analysis of such streaming data at a large scale. Application areas of Data Stream Mining techniques include real-time telecommunication data, telemetric data from large industrial plants, credit card transactions, social media data, Smart Cities, IoT, etc. Some applications allow the data to be processed modelled and analysed in batches by traditional Data Mining approaches. However, others require the model building and analytics to take place in real-time as soon as new data becomes available i.e. to accommodate infinite streams and fast changing concepts in the data. This talk discusses challenges, barriers, opportunities and recent and current research approaches towards innovative solutions in Data Stream Mining.