Foundation concept: Artificial intelligence.

Definition: Machine learning is the ability of programs to make inferences and expand their understanding of the domain in which they operate.

Overview

The term Machine learning is usually associated with automated learning or Self-programming capabilities rather than the production of solutions from static knowledge bases. Mechanisms for learning under current research include learning by knowledge acquisition, learning by examples, neural networks, case-based reasoning (CBR), genetic algorithms, and learning by discovery.

Knowledge acquisition is a vital component of machine learning and is the mechanism through which a system gathers knowledge, which then has to be represented internally to provide the basis for inference. Major challenges to researchers into machine learning lie in each part of this process: how to capture knowledge of different types (from pictures, speech, text, diagrams), how to represent each of these knowledge types, how to integrate them all, and, of course, how to infer useful new knowledge from old. In light of the multitude of different types of knowledge that systems need to acquire and be able to infer from, many techniques have been developed to assist in this process, including computational learning theory, explanation-based learning, delayed reinforcement learning, temporal difference learning, and using version spaces for learning.

One of the key problems still facing researchers is that of common sense reasoning; whereas a child having once touched a hot-plate can extrapolate that anything glowing red is potentially dangerous, this ability to generalize an experience is very difficult for programmers to build into computer systems, because every generalization is context dependent and thus variable in nature.