Online stochastic optimization without distributions

摘要: This paper considers online stochastic scheduling problems where time constraints severely limit the number of optimizations which can be performed at decision and/or in between decisions. Prior research has demonstrated that, whenever a distribution inputs is available for sampling, stochatic algorithms may produce significant improvements solution quality over oblivious approaches. However, availability an input distribution, although reasonable many contexts, too strong requirement variety applications. broadens applicability by relaxing this and using machine learning techniques or historical data instead. In particular, it shows that engineered to learn online, when its underlying structure not available. Moreover, presents idea sampling provides simple effective way leverage continuous periodic optimization. Experimental results on packet vehicle routing indicate potential scheduling.