arXiv cs.CL· Hyojung Han, Jongmin Kim, Seung-Hun Jeon·· 3 天前AI 评分15
反事实门控支撑分解（CGSD）：在训练数据过滤中分离选择与再分配
Counterfactual Gate-Support Decomposition: Separating Selection from Redistribution in Training-Data Filtering
AI 导读
研究者提出反事实门控支撑分解（CGSD），一种四臂、预算匹配的实验设计，将门控效应拆解为支撑内选择、支撑间再分配及二者交互项。在韩英与韩日检索任务上，支撑内选择是唯一超出校准零假设区间的成分（+29.7 与 +5.6 保留项，30/30 正向重复），模型置信度阈值门控也复现该模式（+27.3 项，9/9 正向）。该方法可避免将支撑再分配误读为样本级门控质量。
正文 · 原文
Abstract:A training-data quality gate changes both which examples survive within regions of the input space and how training mass is distributed across those regions, so a gate-versus-random comparison does not isolate these mechanisms. We introduce Counterfactual Gate-Support Decomposition (CGSD), a four-arm, budget-matched design that crosses gate eligibility with support distribution and decomposes the observed gate effect into within-support selection, between-support redistribution, and their interaction. The practical constraint is finite-pool realizability: we characterize mass and token-assembly conditions under which the required arms can be constructed, together with a support-extinction case in which the ordinary design is undefined. On Korean-English and Korean-Japanese retrieval, the mean within-support selection component is the only one that exceeds its calibrated null band (+29.7 and +5.6 holdout items; 30/30 positive replicates), decisively in Korean-English and marginally in Korean-Japanese; redistribution and interaction remain unresolved at the available power. The same pattern holds for a second, unrelated gate family -- a model-confidence threshold rather than a deterministic verifier -- judged against its own null (+27.3 items, 9/9 positive), so the result is not a property of one gate's mechanism. On a mathematics gate, zero-eligibility supports leave the ordinary design undefined at every budget, while the extended design is constructible only inside narrow budget bands that an exhaustive scan locates and that the natural operating point misses -- the assembly condition separating a property of the pool from a limitation of the packer. CGSD therefore returns a decomposition, a constructible-but-underpowered design, or an explicit non-constructibility result, and says which; it prevents support redistribution from being silently interpreted as sample-level gate quality.
| Comments: | 11 pages, 2 tables, 2 figures |
| Subjects: | Machine Learning (cs.LG); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.19343 [cs.LG] |
| (or arXiv:2610.19343v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.19343 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jong Min Kim [view email]
[v1]
Wed, 14 Oct 2026 00:36:40 UTC (53 KB)
来源：arXiv cs.CL · arxiv.org