Machine Learning: Definition, Development History, and Algorithm Classification

1. Definition

Machine learning is a multidisciplinary field that encompasses knowledge of probability theory, statistics, approximation theory, and complex algorithms. It uses computers as tools to simulate human learning in real-time and aims to effectively improve learning efficiency by structuring existing knowledge.

There are several definitions of machine learning:

(1) Machine learning is a science of artificial intelligence, where the main research focus is on artificial intelligence itself, particularly on how to improve the performance of specific algorithms through experiential learning.

(2) Machine learning is the study of computer algorithms that can improve automatically through experience.

(3) Machine learning uses data or past experiences to optimize the performance criteria of computer programs.

2. Development History

Machine learning has actually existed for decades or can be considered to have existed for centuries. Tracing back to the 17th century, the derivation of the least squares method by Bayes and Laplace and the Markov chain are foundational tools widely used in machine learning. From 1950 (when Alan Turing proposed the establishment of a learning machine) to the early 2000s (with practical applications of deep learning and recent advancements, such as AlexNet in 2012), machine learning has made significant progress.

Since the study of machine learning began in the 1950s, the research approaches and goals have varied over different periods, which can be divided into four stages.

The first stage: From the mid-1950s to the mid-1960s, this period mainly studied “learning with or without knowledge.” This type of method primarily researched the execution capabilities of systems. During this time, the system’s feedback data was detected by altering the machine’s environment and its corresponding performance parameters. It is akin to providing the system with a program, and by changing its degrees of freedom, the system would be influenced by the program to alter its organization, ultimately selecting an optimal environment for survival. The most representative research of this period is Samuet’s chess program. However, this method of machine learning was still far from meeting human needs.

The second stage: From the mid-1960s to the mid-1970s, this period mainly studied embedding knowledge from various fields into systems, aiming to simulate the human learning process through machines. This stage also utilized knowledge of graph structures and logical structures for system descriptions. During this research phase, various symbols were used to represent machine language, and researchers realized that learning is a long-term process. Deeper knowledge could not be learned from this system environment, leading researchers to incorporate knowledge from various experts into the system, which proved effective in practice. Representative works of this stage include the structural learning system methods by Hayes-Roth and Winson.

The third stage: From the mid-1970s to the mid-1980s, known as the revival period. During this time, the focus shifted from learning single concepts to learning multiple concepts, exploring different learning strategies and methods. This stage began to integrate learning systems with various applications, achieving significant success. The demand for knowledge acquisition in expert systems greatly stimulated research and development in machine learning. After the first expert learning system emerged, example-based inductive learning systems became mainstream, with automatic knowledge acquisition becoming the research goal of machine learning applications. In 1980, the first International Conference on Machine Learning was held at Carnegie Mellon University (CMU), marking the global rise of machine learning research. Subsequently, machine learning began to see extensive applications. In 1984, the second volume of the Machine Learning anthology, co-authored by over 20 artificial intelligence experts including Simon, was published, and the international journal Machine Learning was launched, further demonstrating the rapid development trend of machine learning. Representative works of this stage include Mostow’s guided learning, Lenat’s mathematical concept discovery program, and Langley’s BACON program and its improved versions.

The fourth stage: The mid-1980s is the latest stage of machine learning. The characteristics of machine learning during this period include:

(1) Machine learning has become a new discipline, integrating psychology, biology, neurophysiology, mathematics, automation, and computer science to form the theoretical foundation of machine learning.

(2) Various learning methods are being integrated, and diverse ensemble learning system research is emerging.

(3) A unified perspective on various foundational issues related to artificial intelligence is forming.

(4) The application range of various learning methods is continuously expanding, with some research outcomes being transformed into products.

(5) Academic activities related to machine learning are unprecedentedly active.

3. Algorithm Classification

Machine learning algorithms can be classified into three types: Supervised Learning, Unsupervised Learning, and Reinforcement Learning.

Supervised Learning involves training models using labeled data, primarily addressing regression and classification problems.

Unsupervised Learning involves training models using unlabeled data, primarily addressing clustering and association problems.

Reinforcement Learning describes and solves the problem of agents maximizing rewards or achieving specific goals through learning strategies while interacting with their environment.

In addition, combining supervised and unsupervised learning leads to Semi-Supervised Learning, which is a form of weakly supervised learning.