部分可观察随机系统的黑盒测试活性性质
Black-box Testing Liveness Properties of Partially Observable Stochastic Systems
AI总结:
研究部分可观察随机系统的黑盒测试活性性质,设计重启策略以有限次重启发现规范违反,改进了需完全可观察的先前工作并获得渐近最优期望步数上界。
摘要:
我们研究随机系统和任意 $ω$-正则规范的黑盒测试,明确包含活性性质。给定一个有限状态概率系统,我们只能从初始状态执行它。我们没有关于可达状态数量或概率的信息;此外,我们只能部分观察状态。我们唯一能采取的操作是重启系统。我们设计了重启策略保证,如果规范以非零概率被违反,那么 w.p.1(依概率1)重启次数是有限的,且在最后一次重启后执行的无穷运行违反了规范。这改进了以前需要完全可观察性的工作。我们获得了直到最后一次重启的期望步数的渐近最优上界。我们在多个基准上进行实验,表明我们的策略能够在比以前工作考虑的大得多的 Markov 链中发现违反情况。
英文摘要:
We study black-box testing for stochastic systems and arbitrary $ω$-regular specifications, explicitly including liveness properties. We are given a finite-state probabilistic system that we can only execute from the initial state. We have no information on the number of reachable states, or on the probabilities; further, we can only partially observe the states. The only action we can take is to restart the system. We design restart strategies guaranteeing that, if the specification is violated with non-zero probability, then w.p.1 the number of restarts is finite, and the infinite run executed after the last restart violates the specification. This improves on previous work that required full observability. We obtain asymptotically optimal upper bounds on the expected number of steps until the last restart. We conduct experiments on a number of benchmarks, and show that our strategies allow one to find violations in Markov chains much larger than the ones considered in previous work.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。