DSPPs achieve the best in-distribution calibration, but deep ensembles are more robust to synthetic feature shifts than either deep GP method.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Evaluating Uncertainty in Deep Gaussian Processes
DSPPs achieve the best in-distribution calibration, but deep ensembles are more robust to synthetic feature shifts than either deep GP method.