Subspace clustering
Subspace clustering Feng Zhou BCMI 2007.5.30
Outline • Introduction • Bottom Up Search Methods – CLIQUE – DOC
• Top‐down Search Methods – ORCLUS – PROCLUS
Introduction (SIGKDD 2004, ASU) • Flaws of Traditional Methods – Curse of Dimensions
– Relevance of clusters and dimensions
Introduction II • Goals – Number of clusters (Usually specified by user) – The cluster each item belongs – The dimensions each cluster exists in
• Hierarchy of existing methods
CLIQUE (SIGMOD 1998, IBM Almaden Research Center)
• Features – Axis‐parallel dimension – Density‐based
• Denotations – Unit – Region
CLIQUE II Algorithm • Identification of subspaces that contain clusters. – Using an Apriori‐style algorithm to find dense units. • Candidates of k dimensional dense units
Outline • Introduction • Bottom Up Search Methods – CLIQUE – DOC
• Top‐down Search Methods – ORCLUS – PROCLUS
Introduction (SIGKDD 2004, ASU) • Flaws of Traditional Methods – Curse of Dimensions
– Relevance of clusters and dimensions
Introduction II • Goals – Number of clusters (Usually specified by user) – The cluster each item belongs – The dimensions each cluster exists in
• Hierarchy of existing methods
CLIQUE (SIGMOD 1998, IBM Almaden Research Center)
• Features – Axis‐parallel dimension – Density‐based
• Denotations – Unit – Region
CLIQUE II Algorithm • Identification of subspaces that contain clusters. – Using an Apriori‐style algorithm to find dense units. • Candidates of k dimensional dense units
