Supervised Learning

lastmod 2019-11-13created 2014-10-256 edits

Supervised Learning is a machine learning method for inferring a function from training data (Training Data). Training data generally contains the attributes of input objects in vector form, and each vector is labeled with the desired result. Among the functions inferred this way, one that outputs continuous values is called regression (Regression), and one that labels which kind of value a given input vector is, is called classification (Classification). The task of a supervised learner (Supervised Learner) is to correctly infer, from the training data, the value to be predicted for given data. To achieve this goal, the learner must be able to generalize, in a "reasonable" way, even to situations that did not appear in the existing training data. In the psychology of humans and animals, the counterpart is concept learning (Concept Learning).

In other words, it can be used when correct answers are given. There are many such problems, such as Regression and Classification.

Problem

Supervised learning can be broadly divided into two kinds depending on the problem type.

Regression problem

First, a Regression problem is one where the expected target value is continuous. For example, apartment price by floor area is a Regression problem: when a target value y_i is expected for a feature x_i, such as a price of 100 million won at 22 pyeong, 120 million won at 23 pyeong, and 140 million won at 24 pyeong, it is called a Regression problem.

Classification problem

A Classification problem is one where the target value is not continuous but falls into a few discrete values. For example, as in "if the age is in the 20s and the tumor size is 2 cm, negative cancer" and "if the age is in the 50s and the tumor size is 3 cm, positive cancer", when the target value expected for features such as age and tumor falls cleanly into a few classes without continuity, such as positive/negative, it is called a Classification problem.

See also

• Machine learning
• Deep learning

Favorite site

• Wikipedia (en) Description of supervised learning
• Samsung Software Membership PLAYGROUND :: 7th cohort Sinchon, Park Jin-sang: Machine learning #1 - Supervised learning
• [Recommended] Self-Supervised Representation Learning ¹ - Self-supervised learning offers a great opportunity to make better use of unlabeled data while learning in a supervised learning manner. This post covers many interesting ideas for self-supervised learning tasks on images, videos, and control problems.

References