Table 3 Comparison of probabilistic network techniques
From: Deep Learning-Driven Data Curation and Model Interpretation for Smart Manufacturing
Technique | Advantage | Disadvantage |
|---|---|---|
Deep ensemble | Highest model diversity; Easy to parallelize | Multi network structure; Most parameter tuning |
MC dropout | Single network structure; Easy to parallelize | Potentially unstable when applied throughout large network |
Multi-head network | Single network structure; Fewest parameter tuning | Lowest model diversity |