Runtime-Assurance for AI

reportActive / Technical Report | Accesssion Number: AD1113747 | Open PDF

Abstract:

Contents include: V and V of AI/ML: Machine Learning Simplified View; V and V Machine Learning Challenges; V and V Approaches for Trained ML; V and V Approach for Evolving ML; Verifying Physics (Control Theory); Analysis of Mission Progress; Drone Experiment; Enforcing Unverified Components; Periodic Execution Must Finish by Deadline; Real-Time Mixed-Trust Computation.

Author(s):
De Niz, Dionisio
Author Organization(s):
Carnegie Mellon University
Funding Organization(s):
Publication Date:
2020 Jan 01
Pagination:
0

Security Markings

RECORD

Collection: TRECMS
Subject Terms
Modernization Areas:
Autonomy
Communities of Interest:
Autonomy
Descriptor(s):
neural networks, software development, machine learning, artificial intelligence, cyber-physical systems, materials, universities, department of defense, equations, guarantees, hybrid systems, quadratic equations, artificial intelligence computing, artificial intelligence software, computers, governments
Keyword(s):
reluplex, verisig
Subject Categories:
Mathematical and Computer Sciences
Update Date:
2020 Dec 17