2017/05/02 by Dmitry B. Rokhlin, Rokhlin, Dmitry B.
Computer Science · Decision Sciences · Engineering · Mathematics · #35K55 #68T05 #68W27 #Advanced Bandit Algorithms Research #Advice (programming) #Applied mathematics #Argument (complex analysis) #Auction Theory and Applications #Computer science #Engineering #Epistemology #FOS: Computer and information sciences #Game Theory and Applications #Limit (mathematics) #Machine Learning (cs.LG) #Machine learning #Mathematical analysis #Mathematical economics #Mathematical optimization #Mathematics #Operations research #Partial differential equation #Philosophy #Regret #Sequence (biology) #Simple (philosophy) #Time limit #Upper and lower bounds #cs.LG #msc:35K55 #msc:68T05 #msc:68W27
paper · pdf · doi:10.48550/arxiv.1705.01091
7 pages
arxiv created 2017/05/02 · openalex publication_date 2017/05/02 · arxiv updated 2017/05/03 · openalex created_date 2022/10/03 · openalex updated_date 2026/08/06
We consider a sequence of repeated prediction games and formally pass to the\nlimit. The supersolutions of the resulting non-linear parabolic partial\ndifferential equation are closely related to the potential functions in the\nsense of N. ,Cesa-Bianci, G. ,Lugosi (2003). Any such supersolution gives an\nupper bound for forecaster's regret and suggests a potential-based prediction\nstrategy, satisfying the Blackwell condition. A conventional upper bound for\nthe worst-case regret is justified by a simple verification argument.\n