Verification of Autonomous Systems: Hyperproperties in Machine Learning

Abstract

Learning-enabled cyber-physical systems (LE-CPS) rely on data-driven learning-enabled components (LECs), such as neural networks (NNs), for tasks ranging across perception, planning, and control to enable autonomy, where such LECs are created with machine learning (ML) methods. While model-based engineering (MBE) has had significant success in providing guarantees for CPS, characterizing the behaviors of LE-CPS given their dependence on LECs remains a significant challenge: in essence, what should LECs do or not do, particularly when composed and integrated into LE-CPSs?

Document Details

Document Type
DoD Grant Award
Publication Date
Jan 21, 2022
Source ID
FA95502210019XX0

Entities

People

  • Taylor T. Johnson

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • Vanderbilt University

Tags

Fields of Study

  • Computer science

Readers

  • Marine Ecological Systems Migration
  • Neural Network Machine Learning.
  • Software Engineering.

Technology Areas

  • AI & ML
  • AI & ML - Autonomous Systems
  • AI & ML - Neural Networks
  • Autonomy
  • Autonomy - Autonomous System Control
  • Cyber