作者: Eyal Gofer
DOI:
关键词:
摘要: This work examines online linear optimization with full information and switching costs (SCs) focuses on regret bounds that depend properties of the loss sequences. The SCs considered are bounded functions a pair decisions, is augmented total SC. We show under general conditions for any normed SC,σ(x,x ' ) = kx−xk, cannot be given only bound Q quadratic variation losses. With an additional � length losses, we prove O( √ + �) Regularized Follow Leader (RFTL). Furthermore, Q) holds RFTL cost kx − xk 2 . By generalizing Shrinking Dartboard algorithm, also expected best expert setting SC, expert. As vanish, all our purely variation. apply results to pricing options in arbitrage-free market proportional transaction costs. In particular, upper price "at money" call options, assuming stock minimum summed gains