Machine learning is a subset of artificial intelligence; neural networks a subset of machine learning. Among neural networks, we have shallow learning (relatively unheard of) as well as deep learning (well known and powerful). Deep learning and random forests (a subset of ensemble learning) account for most of the recent magical breakthrough. 

Machine learning is finally delivering its long prophesied magic! Far from new, it has been an integral part of computer science curricula for decades. It had its ups and downs. The boom did not happen until the recent two or three years. Why?

That’s partly due to the vast availability of 1) data, data and data everywhere, hence the term data science; 2) computing power (notably the GPU) and data storage options everywhere (notably in the cloud). 

Why is machine learning exciting? It is proven to be able to deliver magic to many, many different fields (in order to sound sophisticated researchers call fields of use applications). Examples include predicting sales patterns to aid setting sales strategies; decoding photos, soundclips and thumb prints to identify individuals;  analysing medical scans to suggest diagnosis and treatment.

As a computational scientist venturing into machine learning, the biggest leap for me has been to move beyond algorithm-based, bottom-up, cause-and-effect reasoning. This takes a leap of faith for a physicist.

I no longer teach the computer by designing algorithms. Rather, I feed the computer with lots and lots of data, and leave it to the computer to make out patterns from the data.

Landmark success cases leading recent breakthroughs managed to demonstrate how computers are able to learn certain things better than humans do.