If you think what you see in science fiction movies are not possible, then you are living in the 19th century. In our universe today, science is beginning to coincide with those fictions that were not agreed upon more than a century ago. Artificial intelligence is advancing and while its progress is tremendous, it is dividing into various functional parts of which machine learning belongs.
Machine learning was first used by Samuel Arthur and it is defined as the systematic study of computer models, and statistical algorithms that computers use to carry out specific functions and tasks by using inference and patterns without using specified instructions.
In machine learning, algorithms are being computed to build mathematical models using training or sample data. The model hence helps to make accurate prediction and to make decision without necessarily being programmed to perform the task. Algorithms used in machine learning function in different applications like email filtering, visualization etc., and most especially, where it may be difficult to develop algorithms that will tackle task effectively.
When the concept “machine learning” was coined 1959 by Arthur Samuel, it was based on the premise that computer can learning, more like the way human can also learn. Logically, Tom Mitchell expressed an opinion of machine learning that; computer can be said to learn from experience based on some tasks and measure of performance in those tasks if the performance measure in those tasks improves with experience.
Machine Learning Definition
While this may look too compounded, it simply means that machines can learn, and learn to be better based on what they are been used to do. The illustration given by Tom in terms of the ‘task’ which machine can “Do”, replaced what was used by Alan Turing in his proposal on Machinery and Intelligence based on the question if Machines can “Think”.
From task performance viewpoint that the machine learning definition is based, you need to understand that machine learning tasks have a broad categorical classification. Some of these are; supervised learning, active learning, semi-supervised learning and so on.
Supervised learning involves building a mathematical model algorithm from data containing the input and output desired. A simple supervised model is when for instance you search for a content out of multiple in a cluster of objects with similar characteristics, the algorithm easily separates the objects of your desired characteristic from the rest – although, machine learning has gone far beyond this level as when you surf for a distinctive item, that is exactly what you will get while others are hidden.
Supervised learning makes use of classified and regression algorithms to restrict outputs to limited values. These are the two features that this category works with in presentation of outputs for output presentation.
Another category is unsupervised learning, and that involves building mathematical model algorithm from sample data that has only inputs but not output. With an unsupervised algorithm, structure can be found in data, patterns in data can be discovered, and inputs can be grouped into categories.
In a general view, if there is an algorithm that deals with identifying spam emails, automatically, and because it is a classification algorithm, the output of the task would be a prediction of the email whether it is “spam” or “non-spam”. This is a case of supervised learning in machine learning algorithm function.
So, today, machine learning has advanced beyond simple classification and regression algorithm into complex, more detailed and faster task performance. Today, robots learning algorithms are generating their learning experiences to learn new skills. Guidance mechanisms like active learning, maturation, imitation etc., are being used by robots to aid performances now.
