Amortized Conditional Normalized Maximum Likelihood: Reliable Out of
            Distribution Uncertainty Estimation

                        Aurick Zhou 1 Sergey Levine 1

             Abstract               diction accuracy, they generally do not provide the means
    While deep neural networks provide good per-       to accurately quantify their uncertainty. This is especially
    formance for a range of challenging tasks, cal-     true on out-of-distribution inputs, where deep networks tend
...