Beyond Data Mining: Challenges of interactive machine learning for discovering knowledge in biomedical data

Experts in the life sciences today are confronted with increasingly big data sets, however, the challenge is not the size of this data: Human experts are excellent at pattern recognition in dimensions of 3, the problem is that most of complex data sets are in dimensions much higher than 3, making manual analysis often impossible. Machine learning algorithms may be of help here, and a best practice today is demonstrated by autonomous vehicles ("Google car"). However, in such complex domains as biomedicine, we are confronted with unknown, uncertain, probabilistic, weakly structured data sets. The application of fully automatic machine learning algorithms endangers the modelling of artifacts. The challenge is to bring the human-in-the-machine-learning-loop, hence to make use of the best of two worlds, aiming to support human intelligence with machine intelligence. The long term goal of our research is to contribute towards cognitive computing systems, that learn and interact naturally with experts-in-the-loop to extend what neither a human nor a computer could do on its own.