Online Self-Concordant and Relatively Smooth Minimization, With Applications to Online Portfolio Selection and Learning Quantum States
share
Consider an online convex optimization problem where the loss functions are self-concordant barriers, smooth relative to a convex function h, and possibly non-Lipschitz. We analyze the regret of online mirror descent with h. Then, based on the result, we prove the following in a unified manner. Denote by T the time horizon and d the parameter dimension. 1. For online portfolio selection, the regret of EG, a variant of exponentiated gradient due to Helmbold et al., is Õ ( T^2/3 d^1/3 ) when T > 4 d / log d. This improves on the original Õ ( T^3/4 d^1/2 ) regret bound for EG. 2. For online portfolio selection, the regret of online mirror descent with the logarithmic barrier is Õ(√(T d)). The regret bound is the same as that of Soft-Bayes due to Orseau et al. up to logarithmic terms. 3. For online learning quantum states with the logarithmic loss, the regret of online mirror descent with the log-determinant function is also Õ ( √(T d) ). Its per-iteration time is shorter than all existing algorithms we know.
READ FULL TEXT