Generating Multiple Hypotheses in Non-negative Matrix Factorization and Related Linear Models

Abstract

This proposal describes several novel directions for discovering multiple optima if they exist. As such,it is a fundamentally different tack than prior work in non-negative matrix factorization which outlines assumptions under which the discovered solutions are unique; instead of making assumptions—which may or may not be true—this proposal focuses on providing the domain expert a set of hypotheses to explore patterns in complex data.

Document Details

Document Type
DoD Grant Award
Publication Date
May 02, 2017
Source ID
FA95501710155

Entities

People

  • Finale Doshi-velez

Organizations

  • Air Force Office of Scientific Research
  • President and Fellows of Harvard College
  • United States Air Force

Tags

Readers

  • Distributed Systems and Data Platform Development
  • Linear Algebra