A KL Certificate for Best-of-$N$ Reranking in Language-Model Inference
IEEE Signal Processing Letters · 2026 · 33 · 3307-3311
DOI: 10.1109/lsp.2026.3720982 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Signal Processing Letters, 2026. Verfasst von Yutong Zhang und 1 weiteren Autor:innen. Die Arbeit nennt 11 Literaturangaben.
Abstract (Originalsprache)
Best-of-$N$ reranking draws independent candidates from a reference policy and selects the response maximal under a fixed, sample-independent strict total order on outcomes. The selected law may differ substantially from the reference in Kullback–Leibler divergence. Prior work introduced a bounded statistic depending only on the accepted response's reference mass and conjectured that its expectation upper-bounds this divergence. This letter proves the conjecture for every finite ordered distribution. The proof applies to the full positive cumulative-distribution-function power family, not only integer sample counts. It combines a strictly monotone binary gauge, a top-atom chain-rule recursion, and induction, and yields an exact nonnegative slack decomposition and a quantitative tightness bound. A beta-quantile representation identifies the universal divergence cap. We also treat reference-preserving reward ties, provide deterministic high-precision illustrations, derive clipped fixed-sample confidence bounds, and specify stable evaluation and exact probability-logging requirements.
Zitieren
APA
Zhang, Y., & Yang, Y. (2026). A KL Certificate for Best-of-$N$ Reranking in Language-Model Inference. IEEE Signal Processing Letters, 33, 3307-3311. https://doi.org/10.1109/lsp.2026.3720982
Vancouver
Zhang Y, Yang Y. A KL Certificate for Best-of-$N$ Reranking in Language-Model Inference. IEEE Signal Processing Letters. 2026;33:3307-3311. doi:10.1109/lsp.2026.3720982
BibTeX
@article{Zhang2026,
author = {Zhang, Yutong and Yang, Yaoran},
title = {A KL Certificate for Best-of-$N$ Reranking in Language-Model Inference},
journal = {IEEE Signal Processing Letters},
year = {2026},
volume = {33},
pages = {3307--3311},
doi = {10.1109/lsp.2026.3720982}
}
EUVIMED