Machine learning has become increasingly popular in the last couple of years. Although initially a mostly academic field of research it is now widely being adopted by businesses all over the world, promising smart systems that can outperform humans in both speed and precision. Many of its advantages seem new and groundbreaking. In turn, it also seems to bring a whole new set of potential problems.
In this talk, I want to take a look at some of the fundamental mathematical problems that machine learning is trying to solve, demonstrating that they are in fact much older than the field of AI itself, and that the seemingly new issues arising from machine learning today have actually been known for decades. Furthermore, I will show how the process of learning is defined in some of the most commonly used algorithms and how this relates to the interpretation of their results.