Machine Learning Methods: Discriminative and Generative Models, Semi‑Supervised Learning, and GAN‑Based Classification

Machine Learning Methods

Discriminative models directly model the conditional probability p(y|x) such as Support Vector Machines and logistic regression.

Generative models model the joint distribution p(x,y) including Naïve Bayes, Hidden Markov Models, and Gaussian Mixture Models.

Background and Significance

In practice, labeled samples are scarce and annotating data consumes substantial resources; deep learning further increases the demand for labeled data. Semi‑supervised learning leverages a small amount of labeled data together with abundant unlabeled data to train reliable models.

Semi‑Supervised Learning

To use unlabeled samples, a hypothesis is required that labeled and unlabeled data share an underlying distribution, implying the dataset follows a regular, learnable pattern; otherwise, even abundant labeled data cannot yield a reliable model.

Semi‑Supervised Classification Algorithms

Generative‑based methods assume all data, labeled or not, are generated by the same latent model.

GAN‑Based Semi‑Supervised Classification Algorithm

The article references external resources that illustrate GAN‑based semi‑supervised classification approaches.