Driven by emerging technologies, like the Internet of Things (IoT), cyber-physical systems (CPS), and the so-called Industry 4.0, more data than ever is produced and collected in today’s industries. Often measured by sensors, this data captures the behaviour of a system. By detecting patterns, valuable insights and future predictions can be derived from the collected data.To process the rising data flood, automatic methods are key. Machine learning algorithms, like neural networks, have recently proven their suitability to extract unknown insights from massive datasets.However, to unleash their full potential, data scientists need to evaluate various learning algorithms and tune their parameters, based on their assumptions, against concrete problems and training datasets. This is known as inductive bias or learning bias. While these choices might be satisfactory today, they can be wrong tomorrow, when the behaviour of a system changes.Learning algorithms are designed to converge, which naturally creates a resistance to learn changes and adapt to new situations. While this works impressively well for tasks like image recognition, where more or less static concepts, e.g. the shape of a cat, need to be recognized. However, this makes defining appropriate machine learning algorithms and their parameters at design time extremely challenging for domains like IoT and Industry 4.0, which need to handle continuously evolving data.We propose a continuous meta learning approach that aims at constantly challenging learning algorithms and their parameters against alternatives to dynamically adapt the used learning strategy. We plan to apply reinforcement learning to allow systems to autonomously adapt and improve their learning ability, based on context models developed during my formerly funded AFR PPP PhD (6816126). This work will be a collaboration between DataThings, a company dedicated to develop intelligent software systems and the SnT/UL, which conducts high-quality research in data analytics.