arXiv cs.AI· Jian Gao, Xiao Zhang, Xun Zhu, Miao Li, Ji Wu·· 2 天前AI 评分24
任务级自然语言先验作为低资源 LLM 训练的学习信号
Task-Level Natural Language Priors as Learning Signals for Low-Resource LLM Training
AI 导读
研究提出 Prior-Guided Tuning（PGT）方法，其实现 Contrastive Prior Steering（CPS）在保留监督目标的同时加入正负先验条件损失，方向相反地优化。
正文 · 原文
Abstract:Large language models (LLMs) often struggle when low-resource training data are ambiguous or incomplete. Task-level natural-language priors can provide useful guidance in such settings, but existing approaches usually treat these priors as input context rather than as learning signals during training. We study how task-level natural-language priors can supply guidance missing from these data through Prior-Guided Tuning (PGT). Its implementation, Contrastive Prior Steering (CPS), retains the supervised objective and adds positive- and negative-prior-conditioned losses with opposing optimization directions. Across AmbiMath, Diagnostic Anchoring, and Jigsaw, CPS achieves near-perfect exact-match accuracy on the synthetic task, improves accuracy under misleading diagnostic suggestions by 8.2 percentage points, and improves average subgroup Macro F1 on Jigsaw by 10.7 percentage points over plain fine-tuning, averaged over the two evaluated models. On Jigsaw, with one tenth of the experimental training data, CPS slightly exceeds full-data plain fine-tuning. These results show how reusable task-level language can complement limited labeled examples without requiring instance-level rationales or preference pairs.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2609.02244 [cs.AI] |
| (or arXiv:2609.02244v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2609.02244 arXiv-issued DOI via DataCite |
Submission history
From: Jian Gao [view email]
[v1]
Wed, 2 Sep 2026 07:49:56 UTC (520 KB)
[v2]
Thu, 15 Oct 2026 02:43:49 UTC (277 KB)
来源：arXiv cs.AI · arxiv.org