Efficient Minimax Strategies for Square Loss Games

Neural Information Processing Systems 

We consider online prediction problems where the loss between the prediction and the outcome is measured by the squared Euclidean distance and its generalization, the squared Mahalanobis distance.