Verifying Sensor-Noise Robustness of Reinforcement Learning

Abstract

Neural networks are often vulnerable to adversarial examples, where tiny changes to the inputs can alter the network’s output. In autonomous systems involving sensors, an alarming implication of this is that system correctness could depend on the amount of sensor noise observed at runtime. How much noise does it take to make a system unsafe? Our objective is to create a verification procedure that provides bounds for an autonomous system's observation noise tolerance.

Document Details

Document Type
DoD Grant Award
Publication Date
Jan 21, 2022
Source ID
FA95502210029XX0

Entities

People

  • Stanley Bak

Organizations

  • Air Force Office of Scientific Research
  • Research Foundation for the State University of New York
  • United States Air Force

Tags

Fields of Study

  • Computer science

Readers

  • Educational Psychology
  • Mathematical Modeling and Probability Theory.
  • Neural Network Machine Learning.

Technology Areas

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