Last modified: Mon Aug 22 15:53:00 EDT 2005 by Dean Foster

Statistical Data mining: OUTLINE

Outline: (26 days of class)

• Introduction: The modeling spectrium (1 class)
  • linear regression: the paradign low dimention model
  • nearest neighbor: the paradign infinite dimention model
  • course goals: find a happy middle ground
• Introduction to high dimensional data (intuition isn't a good guide anymore)
• Variable Selection: (4 classes)
  • Bonferroni/Risk inflation
  • FDR/Simes
  • alpha spending, alpha investing
  • Information theory
• Wavelets: Such pretty pictures!
• Loss functions (1 class)
  • Classification loss
  • proper scoring rules: mixtures of clasification loss
  • KL divergence, quadratic losses
• Lasso (1 classes) l1-priors
• regularization (1 class) l2 priors
• Computing p-values (3 classes)
  • White estimator / GEEs
  • Bennett's bound and other "tight" probabilistic bounds
• variable creation (4 classes)
  • interactions
  • missing data
  • RKHS
  • PCA
  • tree stubs
• Searching for natural kinds (2 classes)
  • clustering
  • RBF
• Text data (2 classes)
  • The wikipedia
  • bag of words model
  • Using other peoples parses
• Inductive Logic Programming (2 classes)
  • citation graphs (i.e. links in wikipedia / www)
  • expanding activation
• Non-regression methods (4 classes)
  • SVM
  • Trees
  • boosting
  • comparison to regression