Tensor methods for large-scale learning

Abstract

Learning hidden structures in n-ary relationships inherent in massive amounts of data that are typically generated could lead to state space explosion if handled in a straight forward way. For this STIR work, the PI will address questions of theoretical guarantees that can be obtained with tensors as intentional representation of the data sets. The PI will investigate how to build intentional tensor representations of the n-ary relationships with the proviso that the data structure is built only if needed. The goal is to then construct lower dimensional subspace, where the actual data resides, which then can be used for Machine Learning applications. The approach that the PI will take in showing efficacy of the proposed tensor representation is to show that the errors can be bounded. The PI will also start implementation of the tensor representations to support her theoretical guarantees.

Document Details

Document Type
DoD Grant Award
Publication Date
Jan 12, 2017
Source ID
W911NF1610134

Entities

People

  • Anima Anandkumar

Organizations

  • Army Contracting Command
  • United States Army
  • University of California, Irvine

Tags

Fields of Study

  • Computer science

Readers

  • Educational Psychology
  • Linear Algebra
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - Machine Learning Algorithms
  • Space