Introduction to machine learning with an ALNBench example

Did you know that computers can learn? Yes it's true, learning is a very useful capability of computers.

Usually we think of everything computers do as being embodied in programs that people write, but some problems are too hard to solve directly by writing a program. Such tasks often involve inputs coming from sensors. Some computers have to deal with input data in the form of images, sounds, and imprecise measurements. These data can take many forms that really rep resent the same thing. For example think of all the images that represent a handwritten letter "A". Writing a program to deal with all the myriad possibilities of inputs would be a monumental task. Another issue with writing a special program to solve a problem is often the cost, or the time it would take.

One solution is to use "machine learning", which interacts with the real data of the problem, together with information about the desirability of actions that could be taken. During a period of training, the computer itself can how to solve the problem; which amounts to writing a special program by itself. If we knew better how humans learn, we could point to changes in the brain that form the brain's program for solving a problem.