Subliminal learning transfers preferences in vision-language models

作者 PulseAugur 编辑部 · [1 个来源] · 2026-11-09 05:00

研究人员调查了视觉语言模型（VLMs）中的潜意识学习现象，即教师模型通过不相关的训练数据将隐藏的偏好传递给学生模型。涉及不同大小（20亿至70亿参数）的VLMs的实验表明，即使训练任务侧重于几何形状，动物或面部属性的偏见也可以从教师模型转移到学生模型。这种转移导致学生模型更频繁地选择目标属性，在某些情况下差异高达约10个百分点。此外，研究发现，有偏见的学生模型的解释比对照组学生模型更能模仿教师的措辞，表明其偏好和解释风格都发生了变化。 AI

影响揭示了隐藏偏见在VLMs中转移的潜在可能性，影响了训练数据策展和模型可解释性。

排序理由该集群包含一篇详细介绍模型行为新发现的研究论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CL 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

Subliminal learning transfers preferences in vision-language models

本文如何被排名

Signal score: 16 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇详细介绍模型行为新发现的研究论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, model release; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.CL TIER_1 English(EN) · Arman Hatami, Romina Aalishah, Ilya E. Monosov · 2026-11-09 05:00

Vision-Language 模型中的潜意识学习

arXiv:2611.08538v1 Announce Type: cross Abstract: Subliminal learning is a phenomenon in large language models (LLMs) where a teacher passes a hidden preference to a student through training data unrelated to that preference. We study whether subliminal learning also occurs in vi…

本文如何被排名

报道来源 [1]

Vision-Language 模型中的潜意识学习

相关实体

相关话题