{"paper":{"title":"GateKD: Confidence-Gated Closed-Loop Distillation for Robust Reasoning","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"GateKD uses teacher confidence to gate distillation of reasoning steps from large models to smaller ones, creating a closed-loop process that reduces errors.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Kasidit Sermsri, Teerapong Panboonyuen","submitted_at":"2026-05-13T08:04:46Z","abstract_excerpt":"Distilling multi-step reasoning abilities from large language models (LLMs) into compact student models remains challenging due to noisy rationales, hallucinated supervision, and static teacher-student interactions. Existing reasoning distillation methods, including mentor-based approaches, predominantly operate in an open-loop manner, implicitly assuming uniform teacher reliability and consequently propagating erroneous intermediate reasoning. We propose GateKD, a confidence-gated closed-loop distillation framework that enables robust reasoning transfer by treating the teacher as a dynamic ga"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"GateKD consistently outperforms strong open-loop distillation baselines. Notably, GateKD yields substantial gains in logical and symbolic reasoning, remains robust under low-resource distillation settings, and shows clear performance degradation when any gating component is removed.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That teacher confidence scores reliably indicate the correctness of intermediate reasoning steps and that gating on them does not systematically exclude valid but low-confidence reasoning paths or introduce new selection biases.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GateKD introduces a closed-loop distillation framework that uses teacher confidence to gate soft supervision, hidden-state alignment, and attention transfer, outperforming open-loop baselines on reasoning benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"GateKD uses teacher confidence to gate distillation of reasoning steps from large models to smaller ones, creating a closed-loop process that reduces errors.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b27e53a85e52a68f82caa63657efc910c4070eb59e79fa4066bcf52877ad06e2"},"source":{"id":"2605.13136","kind":"arxiv","version":1},"verdict":{"id":"06a81d5a-cc8f-4aab-80a4-ffa9c5633377","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:34:04.538199Z","strongest_claim":"GateKD consistently outperforms strong open-loop distillation baselines. Notably, GateKD yields substantial gains in logical and symbolic reasoning, remains robust under low-resource distillation settings, and shows clear performance degradation when any gating component is removed.","one_line_summary":"GateKD introduces a closed-loop distillation framework that uses teacher confidence to gate soft supervision, hidden-state alignment, and attention transfer, outperforming open-loop baselines on reasoning benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That teacher confidence scores reliably indicate the correctness of intermediate reasoning steps and that gating on them does not systematically exclude valid but low-confidence reasoning paths or introduce new selection biases.","pith_extraction_headline":"GateKD uses teacher confidence to gate distillation of reasoning steps from large models to smaller ones, creating a closed-loop process that reduces errors."},"references":{"count":20,"sample":[{"doi":"","year":2023,"title":"arXiv preprint , volume=","work_id":"dc288403-5d0d-4749-9308-c3c372c523fb","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Findings of the Association for Computational Linguistics: EMNLP 2024 , pages=","work_id":"ab779dd5-c01c-4557-9729-fc685860b623","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"arXiv preprint , volume=","work_id":"d7d5278c-d572-4057-a571-8f1b9803d3b3","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"arXiv preprint , volume=","work_id":"6b894548-f7d0-4a55-8717-6d8cfe9edd4e","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Proceedings of the 2025 International Natural Language Generation Conference , pages=","work_id":"8a512deb-f3ca-43d8-872d-ac45d7aacfb3","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":20,"snapshot_sha256":"d4b0a0ebd48389903f55f9a00c51242cf3dd8ec91a719d40947d176e0f5d0df5","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}