PAD captures class-specific distributional properties of DNN hidden layer activations and defines deviation measures for uncertainty estimation, competitive inference accuracy, and out-of-distribution sample isolation on image benchmarks.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2019 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Prior Activation Distribution (PAD): A Versatile Representation to Utilize DNN Hidden Units
PAD captures class-specific distributional properties of DNN hidden layer activations and defines deviation measures for uncertainty estimation, competitive inference accuracy, and out-of-distribution sample isolation on image benchmarks.