A Unified Per-Token Gating Family for On-Policy Distillation: FKL/RKL Mixing with Multi-Channel and Bias Coefficients

• URL: http://arxiv.org/abs/2609.11768v1
• Date: Thu, 10 Sep 2026 16:22:35 GMT
• Title: A Unified Per-Token Gating Family for On-Policy Distillation: FKL/RKL Mixing with Multi-Channel and Bias Coefficients
• Abstract summary: Per-token gating of forward/reverse KL losses has become a standard technique for on-policy knowledge distillation (OPD)<n> EOPD and ToDi each fix a single gating signal and a single gating direction, and the two have never been compared directly.<n>We introduce a four-coefficient parameterization_t = sigma(a * h_t + b * u(x) + c + d * gap_t) in which direction-aligned proxies of EOPD and ToDi appear as one-dimensional (1D) restrictions.
• Score: 6.033644084569457
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• 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) + c + d * gap_t) in which direction-aligned proxies of EOPD and ToDi appear as one-dimensional (1D) restrictions, and which adds multi-channel composition and an explicit bias as further degrees of freedom. On TweetEval (Barbieri et al., 2020) emotion and hate, with a Qwen3-32B teacher and a Qwen3-4B student, configurations in the full family reach higher accuracy than the matched-magnitude single-channel (entropy-only / gap-only) 1D restrictions in 33 of 36 comparable cells, and a 26-cell mean-match isolation experiment places dynamic gating ahead of effective-KL-matched static baselines in 19 of 26 cells. Because cells share training data, models, and parameter substructure, we report both counts as exploratory aggregate directional evidence rather than as independent hypothesis tests. Targeted three-seed paired replications of the nine headline comparisons singled out by that sweep -- including a third task, offensive -- are directionally consistent, but individually smaller than the single-seed estimates and not significant at n=3. We therefore present the parameterization primarily as a shared coordinate system for comparing per-token gating designs in short-output classification OPD.

Related papers

• Scaling-Score Conformal Prediction for Multi-Target Regression [0.42970700836450476]
  Multi-target regression requires a model to simultaneously predict several related outputs.<n>We propose the scaling-score conformal method, which is model-agnostic (requires only component-wise absolute residuals)
  arXiv  Detail & Related papers  (2026-09-15T12:23:57Z)
• SCOPE-OPSD: Fisher-Conditioned Privileged Subspaces for On-Policy Self-Distillation [22.56867206175605]
  SCOPE-OPSD projects the privileged teacher-student residual onto a frozen rank-64 factor.<n>It reuses the forwards already required by OPSD and adds neither rollouts nor inference-time modules.<n>Across the complete 25/50/75/100-step trajectories for Qwen3-1.7B, 4B, and 8B, Structured is never below Pure OPSD.
  arXiv  Detail & Related papers  (2026-09-11T08:33:52Z)
• Exact Degeneracy Under Balanced k-Shot Sampling:Consequences for Small-Sample Discriminant Analysis on LLM Embeddings [0.0]
  We show that balanced k-shot sampling induces an exact, provable degeneracy in a family of small-sample discriminant estimators.<n>We then evaluate the Kernelgression repaired framework on few-shot text classification on frozen embeddings.
  arXiv  Detail & Related papers  (2026-09-09T08:13:54Z)
• Feature Priming in Online Linear Regression: Sparse-Regret Lower Bounds and a Tight Univariate Rate [47.36630149538994]
  In high-dimensional online prediction, the best predictor may depend on only a few features, so regret should scale with sparsity rather than the ambient dimension.<n>Warmuth and Amid asked at COLT 2023 whether any of three such rules admits a competitive online regret guarantee.<n>Our analysis identifies a common obstruction: cheap nuisance causes the refit to underweight the truly predictive coordinate.
  arXiv  Detail & Related papers  (2026-08-18T09:32:03Z)
• Measuring the Symmetry--Data Exchange Rate [0.2538209532048867]
  We show that a wrong-group control with identical orbit size and matched compute is worse than no constraint.<n>Relative exchange rate beta_diff = 1.28 is consistent in sign and order of magnitude with the theoretical bootstrap 1.0.<n>The methodological contributions -- the relative-rate estimator that cancels the shared-difficulty confound, the wrong-group control, and a pre-specified failure taxonomy -- transfer to any inductive bias whose strength can be parameterised.
  arXiv  Detail & Related papers  (2026-05-31T08:17:40Z)
• The Matching Principle: A Geometric Theory of Loss Functions for Nuisance-Robust Representation Learning [0.5854803320592717]
  We prove optimality in a linear-Gaussian model (Thm. A), necessity of range coverage for any quadratic penalty that zeros deployment drift (Thm. G)<n>Thirteen blocks (ML through Qwen2.5-7B) test matched vs isotropic vs wrong-direction penalties on geometry and deployment drift.<n>We claim a falsifiable design recipe: estimate Sigma_task, match Sigma', run the controls, report task and geometry separately.
  arXiv  Detail & Related papers  (2026-05-21T17:53:28Z)
• Amplifying, Not Learning: Fine-Tuned AI Text Detectors Amplify a Pretrained Direction [51.56484100374058]
  Text detectors amplify a pretrained typicality axis; they do not construct an AI-vs-human boundary.<n>On raw encoders before any task supervision, projecting onto centroid(AI)-centroid(HC3) achieves NYT-vs-HC3 AUROC 0.806/0.944/0.834 across three architectures.<n>On RoBERTa-base, raw projection exceeds fine-tuning; on RoBERTa-base, full fine-tuning reduces discrimination below raw on both fluent-formal populations tested.
  arXiv  Detail & Related papers  (2026-05-20T19:08:38Z)
• Entropy Across the Bridge: Conditional-Marginal Discretization for Flow and Schrödinger Samplers [68.9946633641494]
  Flow matching and Schrdinger bridges define probability, yet their inference grids are usually or inherited from one-endpoint diffusion.<n>We derive a conditional-marginal entropy-rate objective for bridge-aware discretization, separating endpoint-conditioned bridge geometry from marginal flow evolution.<n>On EDM/CIFAR-10, the entropic time-discretization gives the best tested five-step FID (186.3 pm 4.0 versus 200.5 pm 2.9 for linear and 238.0 pm 5.3 for cosine)
  arXiv  Detail & Related papers  (2026-05-15T16:11:10Z)
• The Extrapolation Cliff in On-Policy Distillation of Near-Deterministic Structured Outputs [52.709361620508595]
  ListOPD brings a student to in-domain with an 8B-SFT baseline at one-fifth the parameters.<n>On Amazon Fashion, three pre-registered tests--a fine-grid cliff interval test, and a small-clip cross-prediction--fall within their locked prediction windows--with the small-clip value matching the closed-form prediction below grid resolution.
  arXiv  Detail & Related papers  (2026-05-09T06:48:00Z)
• Departures: Distributional Transport for Single-Cell Perturbation Prediction with Neural Schrödinger Bridges [51.83259180910313]
  A major bottleneck in gene function analysis is the unpaired nature of single-cell data.<n>We approximate Schrdinger Bridge (SB) to tackle unpaired single-cell perturbation data.<n>Our model effectively captures heterogeneous single-cell responses and achieves state-of-the-art performance.
  arXiv  Detail & Related papers  (2025-11-17T08:27:13Z)
• Ensemble Threshold Calibration for Stable Sensitivity Control [0.0]
  We present an end-to-end framework that achieves exact recall with sub-percent variance over tens of millions of geometry pairs.<n>Our approach consistently hits a recall target within a small error, decreases redundant verifications relative to other calibrations, and runs end-to-end on a single TPU v3 core.
  arXiv  Detail & Related papers  (2025-10-02T15:22:28Z)
• Optimal Multi-Distribution Learning [88.3008613028333]
  Multi-distribution learning seeks to learn a shared model that minimizes the worst-case risk across $k$ distinct data distributions.<n>We propose a novel algorithm that yields an varepsilon-optimal randomized hypothesis with a sample complexity on the order of (d+k)/varepsilon2.
  arXiv  Detail & Related papers  (2023-12-08T16:06:29Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.