Skip to main content
eScholarship
Open Access Publications from the University of California

UC San Diego

UC San Diego Electronic Theses and Dissertations bannerUC San Diego

Learning Accurate and Interpretable Decision Rule Sets from Neural Networks

  • Author(s): Qiao, Litao
  • Advisor(s): Lin, Bill
  • et al.
No data is associated with this publication.
Abstract

This thesis proposes a novel way to learn a set of the Boolean rules in disjunctive normal form as an interpretable model for binary classifications. We consider the problem of learning an interpretable decision rule set as training a neural network in a specific architecture and converting each neuron in the network into a set of minimal decision rules. This approach can easily find a set of interpretable decision rules that has similarly high predictive performance as a full-precision fully-connected deep neural network, and the method can balance between accuracy and complexity. Moreover, we prove that the set of decision rules derived from each neuron is minimum, unique, and irredundant with respect to the original neuron. Our method is competitive with several other state-of-the-art rule learning algorithms, even with fewer rules and simpler rule conditions

Main Content

This item is under embargo until July 13, 2021.