A Unified Per-Token Gating Family for On-Policy Distillation: FKL/RKL Mixing with Multi-Channel and Bias Coefficients
Paper recorded by Signals 4 on 2026-09-10 in cs.AI. Abstract reproduced from arXiv; link to the original below.
Published 2026-09-10 on arXiv · recorded by Signals 4 on 2026-09-11
Category: cs.AI · 人工智能 · first seen 2026-09-11
Abstract
Per-token gating of forward/reverse KL losses has become a standard technique for on-policy knowledge distillation (OPD), but existing methods such as EOPD (Jin et al., 2026) and ToDi (Jung et al., 2025) each fix a single gating signal and a single gating direction, and the two have never been compared directly. We introduce a four-coefficient parameterization lambda_t = sigma(a * h_t + b * u(x) +
Read on arXiv →
Cite this page: A Unified Per-Token Gating Family for On-Policy Distillation: FKL/RKL Mixing with Multi-Channel and Bias Coefficients: the #120 most recent of 300 cs.AI papers we have recorded (as of 2026-09-10). Source: Signals 4 (Signals API) — https://data.jiangzhang.ca/signals4/t/papers/a-unified-per-token-gating-family-for-on-policy-distillation-fkl-rkl-mixing-with.html
Free to quote with attribution to “Signals 4 (Signals API)”. Machine-readable:
papers.json
Get 4 AI signals a day by email — free.
Get 4 AI signals a day by email — free