For an artificial vision machine to recognize human faces, for instance, it is typically necessary to previously train it by showing it thousands of images of human faces. However, not only is this process very time-consuming; it is also like a shot in the dark, because there is no control over what the machine learns during its training. Which facial features has it picked up to be able to do its job? No one really knows. The process works, but the machine itself behaves like black box.
Couldn’t training become much faster if it were possible, beforehand, to inject into the machine some knowledge about the relevant features to look for in faces – or, for that matter, any other objects of interest? That’s exactly what the authors of the study asked themselves, and they answered the question affirmatively by resorting to a recent mathematical approach.
“Allowing humans to drive the learning process of learning machines is fundamental to move towards a more intelligible artificial intelligence and reduce the skyrocketing cost in time and resources that current [learning machines] require in order to be trained”. – Mattia Bergomi
Read the full story here