Tata Steel produces steel coils with accurately specified properties, like grade, surface, geometry and mechanical properties. These product characteristics are of critical importance throughout the manufacturing chain for which process parameters and installation settings have to be calibrated, adjusted and controlled. The manufacturing of steel coils is performed in several steps, from slabs at steel casting, towards steel coils by reduction in thickness and forming of mechanical properties via rolling mills. Further additional processing is done for surface treatment of the steel coils, like galvanizing, towards an end-product. This technique is being used to manufacture car doors, hoods, refrigerator doors, razor blades, etc. At each process step value is added to the main characteristics within tighter tolerances. If the characteristics of the steel deviates from specifications, cracks may occur during the stamping process and surface defects may influence the quality of the coatings later in the process. Obviously this will downgrade the steel product. The process to resolve these problems and to ?recalibrate? the rolling and coating installations is very costly in terms of time, fault products, production loss and loss of source material. Like any high-end industrial production process, steel coil production processes at Tata Steel and automotive stamping processes at BMW typically generate huge volumes of highdimensional process control and product quality data, spread over several plants and process stages. The PROMIMOOC-project aims at developing a generic platform for data collection and integration, data-driven modelling and model-based online process control, by which the steel production can be adapted and optimized in real-time. In contrast to more traditional approaches (e.g. six sigma), we propose to combine distributed in-memory database technology, algorithms for nonlinear data mining and automatic model updates and nonlinear multiple objective decision making algorithms into a real-time system for process control. The approach allows for finding optimal process control compromises for conflicting goals such as e.g. product quality, process robustness, and production cost. The scientific challenges are as follows: Online big data analytics: High-volume non standardised data streams generated at different locations need to be combined with a large quantity of historical data to facilitate data mining. Data mining algorithms need to be integrated into a distributed in-memory database to achieve online big data analytics. The consortium will build upon and extend the MonetDB system (www.monetdb.org). Automatic generation of models based on big data: The relationship between process control data and product properties will be modelled through state-of-the-art nonlinear regression algorithms (e.g., support vector machines, random forests, etc.). The automatic generation, comparison, selection and update of such models will be a major task within the project. Multiple objective decision making based on models: The data driven models will be combined with nonlinear global optimization algorithms for finding Pareto-optimal process control settings to achieve best compromises between the conflicting goals, under complex constraints. Evolutionary strategies will be used for this task.

The Dutch-built Low-Frequency Array (LOFAR) is a unique radio telescope that brings together the signals from tens of thousands of antennas spread across the Netherlands and Europe. By removing various bottlenecks in data transport and data processing, we will unlock the full potential of LOFAR to make both sharp and wide-field images of radio waves arriving from outer space. We will study, for example, how stars form over cosmic time and how exoplanets are influenced by their parent star. We will also capture rare explosions from merging stars and study the extremes of the Universe.

Metalearning is the study of principled methods that exploit metaknowledge to obtain efficient models and solutions by adapting machine learning and data mining processes. The related area of automated machine learning (AutoML) is concerned with automating the machine learning processes. While the variety of machine learning and data mining techniques now available can, in principle, provide good model solutions, searching for the most appropriate model may be a difficult task. Metalearning and AutoML provide a methodology that allows systems to become more effective through experience. This book discusses approaches that exploit knowledge concerning machine learning and data mining algorithms, or more complex solutions (ensembles, workflows etc.) that include these algorithms as elements. It presents many current approaches how this knowledge can be reused to select, combine, compose, adapt and configure both algorithms and models to yield faster and better solutions to data mining and data science problems. It can thus help developers to develop systems that can improve themselves through experience. This book is a substantial update of the first edition published in 2009. It includes 18 chapters, more than twice as much as the previous version. This enabled the autors to cover the most relevant topics in more depth and incorporate the overview of recent research in the respective area. The book will be of interest to researchers and graduate students in the areas of machine learning, data mining, data science and artificial intelligence.

This research has developed algorithms for dynamic data analytics that are based on techniques for automatically constructing machine learning pipelines for the task at hand. The approach has been validated on two complementary practical application tasks: The early detection and treatment optimization for Parkinson’s disease, and the cost- and environmentally optimized management of energy for private households with electric vehicles. The project output includes a new online automated machine learning pipeline, new methods for multi-variate time series prediction, and new approaches for early stage Parkinsons disease diagnostics from videos.

The goal of this project was to develop new techniques that allow learning algorithms to be applied to large 3D images. During the project, several techniques were developed and then successfully applied to challenging problems from different application areas. For example, the techniques helped analyze bacteriophages that might be an alternative to antibiotics and analyze the causes of bone fractures.