Figure 2. Full knowledge map of data science [1]

[1] Longbing Cao. Data Science: An Introduction.

What forms the family of data science? What makes data science as a discipline?

There are different types of data inputs, problems, tasks, experiences, evaluation methods, and outputs (e.g., learning a signal or feedback from data or experience) for data science. Accordingly, analytics and learning approaches can be categorized in terms of different foundations, learning tasks, methods, and perspectives.

Figure 1 summarizes the main methods, tasks and/or objectives in machine learning, knowledge discovery, and general data analytics. They can be categorized into groups including:

Fundamentals of analytics and learning: these include theoretical foundations and tools for analytics and learning, in such areas as algebra, numerical computation, set theory, geometry, statistical theory, probability theory, graph theory, and information theory.

Classic research on analytics and learning: which consist of such areas as feature engineering, dimensionality reduction, rule learning, classic neural networks, statistical learning, and evolutionary learning, unsupervised learning, supervised learning, and semi-supervised learning.

Advanced research on analytics and learning: which include such areas as representation learning, Bayesian networks, graphical modeling, reinforcement learning, deep neural networks and deep modeling, transfer learning, non-IID learning, X-analytics, advanced techniques for optimization, inference and regularization, and actionable knowledge discovery.