Supervised and unsupervised learning

Supervised Learning: Learning training samples with concept tags (Classification) to label (Classification) the data outside the training sample set as much as possible. All tags (categories) are known here. Therefore, the ambiguity of the training sample is low.

Unsupervised learning: Learning training samples without conceptual tags (classifications) to discover structural knowledge in the training sample set. Here, all tags (categories) are unknown. Therefore, the training sample has a high level of ambiguity. Clustering is a typical unsupervised learning.

Appendix:

Methods or paradigms in machine learning have many classification systems, such as learning methods, examples, analogy, and analysis. However, in general, what is most researched and considered most useful now is to learn from examples. There are many classification methods for learning from examples. For example, in terms of learning initiative, there can be divided into active learning and passive learning. From the beginning of the training process, it can be divided into urgent learning and inert learning. The most common classification of "learning from examples" is supervised learning, unsupervised learning, and Reinforcement Learning, which are classified by ambiguity. For supervised learning, it learns training samples with a concept label to predict the concept tags of examples outside the training set as far as possible. The concept tags of all training samples are known, so the ambiguity of training samples is the lowest.
For unsupervised learning, it learns training examples without concept tags to discover the hidden structural knowledge in the training examples. The concept mark of the Training example is unknown, so the ambiguity of the training sample is the highest. PairReinforcement LearningFor example, it learns from training examples that do not have a concept mark, but are associated with a delayed reward or utility (which can be considered a concept mark of delay, to obtain a ing from status to action. There is no concept mark here, but the delay reward can be regarded as a concept mark of delay, so the ambiguity of the training sample is between supervised learning and unsupervised learning. Note that supervised learning and unsupervised learning are relative from the very beginning, however, reinforcement learning does not consider the difference between supervised learning and unsupervised learning from the perspective of training sample ambiguity. Therefore, in some early studies, reinforcement learning is considered a special kind of unsupervised learning. In fact, the positioning of Reinforcement Learning is still controversial. Some scholars even think that it is the same level of concept as "learning from Examples. The classification system from the perspective of training sample ambiguity is expected to be expanded in recent years, for example, multi-instance learning) from the aspect of the ambiguity of the training samples, a special new learning framework may enter the system. So far, no new framework has been recognized. In addition,Semi-Supervised Learning(Semi-supervised; Learning) also has some hope. Its obstacle is that ambiguity in semi-supervised learning is not inherent, but artificial, that is, you expect to use unlabeled samples to help you learn labeled samples. This is completely different from supervised learning, unsupervised learning, and reinforcement learning. Semi-supervised learning of human ambiguity is required and useful in solving engineering problems (marking a large number of samples may be very expensive ), however, it may not lead to a major change in methodology or perspective on learning problems. Different classification systems are overlapping. For example, supervised learning methods may be urgent (for example, most Neural Networks and decision trees) or inert (for example, K-Nearest Neighbor ). In addition, the classification system is not absolute, such as the reinforcement learning mentioned above.

Supervised and unsupervised learning