An ensemble of InfoNCE encoders, with their embedding spaces aligned by an unsupervised orthogonal transformation, improves OOD retrieval metrics on MNIST over single encoders.
Url: A representation learning benchmark for transferable uncertainty estimates
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Improving OOD Generalization of Pre-trained Encoders via Aligned Embedding-Space Ensembles
An ensemble of InfoNCE encoders, with their embedding spaces aligned by an unsupervised orthogonal transformation, improves OOD retrieval metrics on MNIST over single encoders.