Feb 28, 2022 · We demonstrate that this approach efficiently distinguishes agents in high-dimensional examples involving simple logistic regression as well as ...

Dyadic sampling can feasibly evaluate joint predictions on high-dimensional real datasets. We evaluate benchmark approaches to Bayesian deep learning and show.

Agent-hyperparameter pairs that perform better on the testbed generally also perform better on real data. This result is statistically significant in both τ = 1 ...

This work introducesdyadic sampling, which focuses on predictive distributions associated with random \textit{pairs} of inputs, which efficiently ...

We demonstrate that this approach efficiently distinguishes agents in high-dimensional examples involving simple logistic regression as well as complex ...

We show that the benefits of domain-specific priors can be pronounced when evaluating high-order joint predictions, even where they are negligible for marginals ...

Usable estimates of predictive uncertainty should (1) cover the true prediction targets with high probability, and (2) discriminate between high- and low- ...

Mar 19, 2024 · In this work, we have analyzed prediction uncertainty of deep neural networks and simple control methods in compound potency prediction and ...

Jul 27, 2021 · Evidential deep learning extends the idea of learning the parameters of a probability distribution further to predict higher-order distributions ...