No-Regret Learning with Unbounded Losses: The Case of Logarithmic Pooling

Neural Information Processing Systems 

We consider the problem of learning the best set of parameters (i.e.