Their training requires constantly increasing volume of samples, and they also do not be able to explain why a particular decision was made. Structural approaches to Machine Learning avoiding these drawbacks exist, the software implementation of one of which is described in the article. This is an English translation of original post by the author.
