God Does Play Dice: Diagnosis and Validation for Autonomous Systems
S Bayana, David R. Owen, Tim Menzies, Susanta Mukhopadhyay · 2004
Randomized algorithms have been known to outperform their deterministic counterpart over a wide range of problems. In this paper, we use randomized techniques for validating and diagnosing autonomous intelligent systems. Such techniques provide efficient approximate solutions to both the diagnosability and the validation problems. In particular, we show the effectiveness of LURCH, a randomized inference engine that we have developed in validating and diagnosing autonomous systems. LURCH uses random search methods that use (1) a fast partial search, (2) a random selection amongst options, and (3) the occasional reset/restart. We have conducted case studies on an optical navigation system, a camera control system and several components a propulsion system all written in a Reactive Model Programming Language (RMPL). 1.