Adaptive Online Estimation of Piecewise Polynomial Trends

We consider the framework of non-stationary stochastic optimization [Besbes et al., 2015] with squared error losses and noisy gradient feedback where the dynamic regret of an online learner against a time varying comparator sequence