How do you prove an autonomous system is safe? From falsification to reachability analysis, from importance sampling to runtime monitoring. The complete validation toolkit.
What is validation? History, societal consequences, the validation framework.
Model building, probability, parameter learning, agent models.
Metrics, composite metrics, temporal logic, reachability specifications.
Direct sampling, disturbances, fuzzing, objective functions, CMA-ES.
Shooting methods, tree search, heuristic search, MCTS, RL.
Rejection sampling, MCMC, probabilistic programming for failures.
Importance sampling, adaptive IS, sequential Monte Carlo, multilevel splitting.
Forward reachability, set propagation, zonotopes, polytopes, LP.
Interval arithmetic, Taylor models, neural network verification.
Graph reachability, SAT, probabilistic reachability, state abstractions.