aka descriptive learning, knowledge discovery

Supervised vs unsupervised

RI Seminar: Yann LeCun : The Next Frontier in AI: Unsupervised Learning

See Machine learning. Given a set of data $D=\{x^{(i)}\}_{i=1}^m$, finding "interesting patterns" in the data.

Intro lec by Andrew Ng – Unsupervised learning

Some of these algorithms are useful for Supervised learning, as a previous step of Model selection, for instance for Feature selection

They are also useful for modeling the prior in a Generative learning approach.

See this video to see how the models are organized

Clustering

K-means algorithm

Community clustering in networks

Mixture model

Generative models

Given a data set of $x$s, build a probabilistic model $P(x)$, for the data.

Vid, can be used for Anomaly detection.

Mixture model

Gaussian mixture model

Factor analysis model

Distribution estimation (aka density estimation)

http://www.iliasdiakonikolas.org/

Often refers to non-parametric generative models.

Dimensionality reduction

Factor analysis model

Principal component analysis

Generalizations: Topological data analysis/Manifold learning. When data lies approximately on an algebraic manifold

Manifold learning

Discovering graph structure

Graphical model learning

Outlier detection

Others

Independent component analysis

Boltzmann machine

Restricted Boltzmann machine

Deep belief network

Autoencoder

Sparse coding model

Useful Optimization algorithm:

EM algorithm

Self-organizing map

Restricted Boltzmann machine

Autoencoder