Machine Learning and It’s Origins
One of the most quoted definitions of Machine Learning is:
The subfield of computer science that “gives computers the ability to learn without being explicitly programmed. (Samuel, 1959)
That is a compact, but also a complete description of the major paradigm shift Machine Learning brings to the world of solving problems, answering questions, and taking decisions with the use of Information Technologies. It implies that we can delegate to a computer the task to make sense out of a dataset “on its own,” without needing humans defining the exact course of calculations and actions, thus without us having understood the true nature of the problem at hand and the path to its solution. That way, the machine uses the data as “learning material” in order to assess and classify new or unseen data under the same context, or predict future values, eventually developing the ability to make decisions or/and define courses of action “on its own.” That human-like ability is described in a definition which was given a few decades later:
A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E. (Mitchell, 1997)
Taking a certain risk of oversimplification, the basic concept can be described as follows: One can take a dataset that he believes (or at least hopes) that contains the necessary information, a truth that cannot be easily discovered but is deemed essential in order to complete a specific task. He acknowledges, at the same time, the fact that, as the volume of data, the number of parameters that take part in the outcome and the complexity of their correlation increase, it becomes increasingly difficult (and at some point impossible) for the human mind to process, come up with a visible and intuitive hypothesis about the hidden patterns and model the acting causalities in order to provide means of accurate assessment and/or prediction. Then, the computer is let to create its universe out of this data, a perception of the reality in the form of multidimensional “hyperspheres”, creating vectors out of every data point, and, by the application of complex mathematic principles, calculate its way to an algorithm, that “understands” the acting causalities and “captures” the underlying patterns, thus becoming capable of being applied as “knowledge” and “experience” towards solving (or helping to solve) related problems.