Semi-unsupervised learning

Semi-supervised Classification

The learning problem is tackled using a large amount of unlabeled data together with a small quantity of labeled data to make a classifier.

The algorithm uses labeled data to modify hypotheses from labeled data.

Since semi-supervised classification requires a moderate human effort for the labeling process, it has a large practical interest.

Semi-supervised Regression

Is a particular case of ss classification in which label are the values of continuos variables, like a financial environment index.

The goal remains the clustering but a few labelled data exist under the form of the so-called must-links and cannot-links.

• Must-link: two data points that must lie in the same cluster
• Cannot-link: two data points that cannot be in the same cluster