Semi-Supervised Learning

Semi-supervised learningIt is a learning technique between supervised learning and unsupervised learning. It uses both labeled and unlabeled samples for learning.

Semi-supervised learning allows the learner to automatically use unlabeled samples to improve learning performance without relying on external interactions.

Two common semi-supervised learning assumptions

One is the "cluster assumption", which assumes that the data has a cluster structure and samples in the same cluster belong to the same category.

The other type is the “manifold assumption”, which assumes that the data is distributed on a manifold structure and that adjacent samples have similar output values.

Brief introduction to semi-supervised learning theory:

Semi-supervised learning has two sample sets, one labeled and one unlabeled.

Lable = { ( xi , yi ) }, Unlabled = { ( xi ) }. And quantitatively, L << U.

1) Using labeled samples alone, we can generate supervised classification algorithms;

2) Using unlabeled samples alone, we can generate unsupervised clustering algorithms;

3) Using both, we hope to add unlabeled samples in 1 to enhance the effect of supervised classification; similarly, we hope to add labeled samples in 2 to enhance the effect of unsupervised clustering.

Generally speaking, semi-supervised learning focuses on adding unlabeled samples to supervised classification algorithms to achieve semi-supervised classification. That is, unlabeled samples are added to 1 to enhance the classification effect.

Classification of Semi-Supervised Learning Algorithms:

1) Self-training algorithm;

2) Generative models

3) SVMs semi-supervised support vector machines;

4) graph-based methods;

5) Multiview learing algorithm.

References

【1】https://blog.csdn.net/ice110956/article/details/13775071

【2】http://blog.sciencenet.cn/blog-242887-309591.html

【3】https://www.zybuluo.com/Team/note/1133211