Generalization is not a universal guarantee: Estimating similarity to training data with an ensemble out-of-distribution metric

Open in new window