Works only with graphs

Iteratively updates cluster centroids

Requires specifying the number of clusters K

Uses random walks

It only works with labeled data

It always requires specifying K beforehand

It can produce a dendrogram

It builds a hierarchy of clusters

Vectors

Edges (links)

Tables

Nodes (vertices)

Requires labeled data

Avoids generating large candidate sets

Uses a compact tree structure

Works only on graphs

Clustering

Classification

Data deletion

Association rule mining

One should adapt to uncertainty rather than attempt perfect prediction.

Risk can be eliminated through sufficient model complexity.

Management responsibility is replaced by automated optimisation.

Balance sheet decisions must reflect both risk and return considerations.