Online Algorithm for Aggregating Experts' Predictions with Unbounded Quadratic Loss

We consider the problem of online aggregation of experts' predictions with the quadratic loss function. At the beginning of each round t = 1,2,...,T, experts n = 1,...,N provide predictions γ The player and experts n = 1,...,N suffer losses h The goal of the player is to minimize the regret, i.e. the difference between the total loss of the player and the Online regression is a popular special case of the considered problem, i.e. γ The problem of online prediction with experts' advice is considered in the game theory [1] and machine learning [3]. Existing aggregating algorithms provide strategies which guarantee a constant upper bound on the regret but assume that the losses are bounded. However, the algorithm requires knowing B beforehand. In this paper, we propose an algorithm for aggregating experts' predictions which does not require a prior knowledge of the upper bound on the losses.