Michal Valko : PhD Thesis

Adaptive Graph-Based Algorithms for Conditional Anomaly Detection and Semi-Supervised Learning

Committee

• Miloš Hauskrecht, PhD (Dissertation Director)
• Liz Marai, PhD
• Diane Litman, PhD
• John Lafferty, PhD (Carnegie Mellon University)

Defense

• Thesis document
• Defense slides (PDF)
• Defense slides (PowerPoint)
• Defense slides (early version)
• Backup slides
• Defense announcement
• Date: Monday, August 1st, 2011
• Time: 10am
• Room: Seminar Room (5317)
• Address: 210 S Bouquet St, Pittsburgh, PA

Proposal

• Proposal slides (PowerPoint)
• Proposal announcement

External Committee Member

• Outside committee member info
• 

Administrative Documents

• Graduate checklist
• PhD checklist
• Graduate instructions
• PowerPoint theme

Demo: Conditional Anomaly Detection

The thesis develops algorithms to detect conditional anomalies: objects that are anomalous only in a specific context. For example: given a basket of apples, can we detect the fruit that doesn't belong?

Normal

Normal

Normal

Anomaly!

Normal

Normal

Anomaly!

Normal

Normal

Anomaly!

Normal

Normal

Mixed

Mixed

Mixed

The graph-based algorithms in this thesis can learn to identify such anomalies using similarity graphs and semi-supervised learning.