Under two axioms (preference-consistency and prevalence-consistency), the only distribution-grounded evaluation for preference learning is h-win rate, and most popular alignment methods can be classified by whether they optimize it.
Multi-class h -consistency bounds
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Preference learning made easy: Everything should be understood through win rate
Under two axioms (preference-consistency and prevalence-consistency), the only distribution-grounded evaluation for preference learning is h-win rate, and most popular alignment methods can be classified by whether they optimize it.