Dimensionality Reduction for Matrix- and Tensor-Coded Data [Part 1]

Dimensionality Reduction for Matrix- and Tensor-Coded Data [Part 1]

Date Posted: 

September 11, 2017

Date Recorded: 

September 5, 2017


Alex Williams
  • Computational Tutorials


Alex Williams, Stanford University

In many scientific domains, data is coded in large tables or higher-dimensional arrays. Compressing these data into smaller, more manageable representations is often critical for extracting scientific insights. This tutorial will cover matrix and tensor factorizations - a large class of dimensionality-reduction methods that includes PCA, non-negative matrix facotrization (NMF), independent components analysis (ICA), and others. We will pay special attention to canonical polyadic (CP) tensor decomposition, which extends PCA to higher-order data arrays.

The first half of the tutorial will cover theoretical concepts and foundations of these methods, many of which are surprisingly recent results. The second half will include hands-on exercises and advice for fitting these models in practice.

After the tutorial, slides and resources will be posted on the computational tutorials stellar page. 
slides, references, and exercises: https://stellar.mit.edu/S/project/bcs-comp-tut/materials.html
videos: http://cbmm.mit.edu/videos?field_video_grouping_tid[0]=781