Verifier-Guided Selection
An inference-time technique that, after generating multiple candidates, uses an independently trained verifier model to score the candidates and select the best output.
Timeline (last 90 days)
Verifier-guided selection means that after generating multiple candidates, an independently trained verifier model scores the candidates and picks the best one.
The PSV framework unifies self-consistency and verifier-guided selection: perturbation generation plays the sampling role, validation-set selection corresponds to verifier screening, and vote aggregation is consistent with self-consistency.
All Facts (2)
Verifier-guided selection means that after generating multiple candidates, an independently trained verifier model scores the candidates and picks the best one.
50%UnverifiedThe PSV framework unifies self-consistency and verifier-guided selection: perturbation generation plays the sampling role, validation-set selection corresponds to verifier screening, and vote aggregation is consistent with self-consistency.
50%