A Universal Error Measure for Input Predictions Applied to Online Graph Problems

Neural Information Processing Systems 

We introduce a novel measure for quantifying the error in input predictions.