Machine Learning

History and Types of Machine Learning

An early definition of machine learning dates back to 1959 and is attributed to Arthur Samuel, who described it as a “field of study that gives computers the ability to learn without being explicitly programmed.” Later, a more formal definition was provided by Tom M. Mitchel, who specified machine learning as the process of a computer program learning from experience, which is associated with specific classes of tasks and can be measured through the use of applicable performance metrics. In general, there are various tasks that can be encountered in machine learning (both in security/surveillance uses and in more generic application areas), such as function modeling, time series prediction, pattern classification/recognition, data clustering, information retrieval, knowledge discovery, and data mining. These tasks are also associated with statistics and the probabilistic properties of the involved data.

While machine learning is strongly related to artificial intelligence and optimization methods and algorithms, there is not a strict definition that could crisply discriminate machine learning from other related disciplines. Some argue that preexisted data sets (inputs and known/desired outputs) are used as training pairs in machine learning, thus guiding this learning-by-experience process, while data mining is an unsupervised/exploratory data analysis concept, where previous experience is not [Page 592]involved. Still, similar attributes and principles are also encountered in unsupervised machine-learning algorithms, where hidden patterns are discovered and clustered without the use of prior class-labeling information. Besides these two cases (supervised and unsupervised learning), reinforcement learning is a kind of subset of supervised learning that is encountered in more dynamic environments, where the whole process is conducted live, with machine interaction simultaneously evolving during training.

...