Contaminated Online Convex Optimization

In the field of online convex optimization, some efficient algorithms have been designed for each of the individual classes of objective functions, e. g. , convex, strongly convex, and exp-concave. However, existing regret analyses, including those of universal algorithms, are limited to cases in which the objective functions in all rounds belong to the same class, and cannot be applied to cases in which the property of objective functions may change in each time step. To address such cases, this paper proposes a new regime which we refer to as contaminated online convex optimization. In the contaminated case, regret is bounded by O (T+k T) when some universal algorithms are used, and bounded by O (T+k) when our proposed algorithms are used, where k represents how contaminated objective functions are. We also present a matching lower bound of (T + k). These are intermediate bounds between a convex case and a strongly convex or exp-concave case.