-
Learning Compressed Transforms with Low Displacement Rank
Introduces a class of LDR matrices where the displacement operators are explicitly learned from data.
-
MetaCleaner: Learning to Hallucinate Clean Representations for Noisy-Labeled Visual Recognition
Proposes a framework which can learn to hallucinate clean representations of noisy labeled data.
-
Combating Label Noise with Abstention
Introduces a loss function that permits abstention during training.