Research

Paper

TESTING February 24, 2026

GATES: Self-Distillation under Privileged Context with Consensus Gating

Authors

Alex Stein, Furong Huang, Tom Goldstein

Abstract

We study self-distillation in settings where supervision is unreliable: there are no ground truth labels, verifiable rewards, or external graders to evaluate answers. We focus on document-grounded question answering with asymmetric context, where a single model serves as both tutor (with access to a relevant source document during training) and student (answering from the question alone at test time). Rather than assuming tutor correctness, we derive supervision online from tutor consensus by sampling multiple document-grounded reasoning traces and using agreement to gate learning. Conditioned on this reliability signal, we distill knowledge through full tutor reasoning trajectories (not just final answers), providing a dense and stable learning signal. Empirically, this consensus-gated trajectory distillation substantially improves transfer to the document-free student. Held-out in-domain accuracy under asymmetric evaluation improves from 46.0\% to 62.0\%, and average (maj@8) accuracy on public document-free math benchmarks improves from 20.2\% to 35.4\%.

Metadata

arXiv ID: 2602.20574
Provider: ARXIV
Primary Category: cs.LG
Published: 2026-02-24
Fetched: 2026-02-25 06:05

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2602.20574v1</id>\n    <title>GATES: Self-Distillation under Privileged Context with Consensus Gating</title>\n    <updated>2026-02-24T05:56:20Z</updated>\n    <link href='https://arxiv.org/abs/2602.20574v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2602.20574v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>We study self-distillation in settings where supervision is unreliable: there are no ground truth labels, verifiable rewards, or external graders to evaluate answers. We focus on document-grounded question answering with asymmetric context, where a single model serves as both tutor (with access to a relevant source document during training) and student (answering from the question alone at test time). Rather than assuming tutor correctness, we derive supervision online from tutor consensus by sampling multiple document-grounded reasoning traces and using agreement to gate learning. Conditioned on this reliability signal, we distill knowledge through full tutor reasoning trajectories (not just final answers), providing a dense and stable learning signal. Empirically, this consensus-gated trajectory distillation substantially improves transfer to the document-free student. Held-out in-domain accuracy under asymmetric evaluation improves from 46.0\\% to 62.0\\%, and average (maj@8) accuracy on public document-free math benchmarks improves from 20.2\\% to 35.4\\%.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.LG'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <published>2026-02-24T05:56:20Z</published>\n    <arxiv:comment>10 Pages of main text with an additional 7 pages of supplementary material</arxiv:comment>\n    <arxiv:primary_category term='cs.LG'/>\n    <author>\n      <name>Alex Stein</name>\n    </author>\n    <author>\n      <name>Furong Huang</name>\n    </author>\n    <author>\n      <name>Tom Goldstein</name>\n    </author>\n  </entry>"
}