Using f-divergence regularizers inside Fenchel-Young losses yields convex losses and fast f-softargmax operators, and the α=1.5 variant matches or beats cross-entropy in the tasks tested.
Robust semi-supervised learning via f-divergence and -r \'e nyi divergence
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
Loss Functions and Operators Generated by f-Divergences
Using f-divergence regularizers inside Fenchel-Young losses yields convex losses and fast f-softargmax operators, and the α=1.5 variant matches or beats cross-entropy in the tasks tested.