notes
code
data
demos

CSSS 594

Multiway data analysis


Lecture and discussion: Tue Thu, 11:30-12:50, PDL C301
  • 11:30-12:30 - Lecture
  • 12:30-12:50 - Discussion

Instructor
  • Peter Hoff
  • C-319 Padelford
  • Office Hours: Mon, Wed 9:30-10:30 or by appointment.
  • pdhoff at uw dot edu

Reading
Calendar
4-2: Introduction ( slides )
4-4: SVD ( notes )
4-9: SVD ( notes )
4-11: Matrix data analysis ( notes, demo )
4-16: Matrix data analysis ( demo 1 , demo 2 )
4-18: Covariance models for matrices ( notes )
4-23: Testing for covariance in networks
4-25: Multilinear operators on arrays ( notes )
4-30: Separable covariance models for arrays ( demo 1 , demo 2 )
5-2: Tensor SVD ( notes )
5-7: Tensor SVD ( notes )
5-9: No class
5-14: Tensor SVD demos ( demo 1 , demo 2 , demo 3 , demo 4 )
5-16: Model-based TSVD ( notes )
5-21: Model averaging and rank selection for TSVD ( notes , demo )
5-23: MBTSVD demos ( demos )
5-28: Separable factor analysis
5-30: Kean and Linbo
6-4: Shizhe and David
6-6: Philip and Theresa
Presentations
  • David: Inference with transposable data (Allen and Tibshirani)
  • Shizhe: A Generalized Least Squares Matrix Decomposition (Allen et al)
  • Kean: Biclustering with the matrix-variate normal distribution
  • Linbo: Bi-cross-validation of the SVD (Owen and Perry, 2009)
  • Theresa: Separable Covariance Models in Spatial Statistics
  • Phillip: Mixtures of Array Normal Distributions, with Applications to Clustering Gene Expression Data

Evaluation
  • Presentation
  • Project

Approximate Course Outline

  1. Review of matrix-variate data analysis
    1. ANOVA and SVD
    2. matrix normal models
  2. Array rank and decompositions
    1. Notions of tensor rank
    2. CANDECOMP/PARAFAC
    3. higher order SVD
    4. theoretical concerns
  3. Mean models and estimation
    1. alternating least squares
    2. penalized likelihoods
    3. Bayesian estimation
  4. Covariance models and estimation
    1. Separable covariance models
    2. Multiway factor models
  5. Applications
    1. Dynamic networks
    2. Multiway relational data
    3. International relations data
    4. Psychometrics
    5. FMRI and imaging data