I recently read a white paper about machine learning written by Tom Mitchell. The article can be found here. This recent paper (July 2006) deals about the discipline of machine learning, its state and position in computer science. Mitchell also writes about current research areas and possible directions for future work. He has a very good definition of machine learning, which is somehow close to data mining: “How can we build computer systems that automatically improve with experience, and what are the fundamental laws that govern all learning processes?“.
He gives good examples of applications that are too complex to be manually coded but can be processed with machine learning. Among research questions, he gives examples such as the use of unlabeled data for supervised learning and the relationships between different learning algorithms. I have in mind, two other points, which are, to my knowledge, still research questions:
- How can we deal with data containing a varying number of parameters?
- Until what point can the machine learning/data mining process be automated?
Out of this well written white paper, Mitchell is the author of a famous book entitled: Machine Learning (1997). Although his book is comprehensive, it is not easy to read.