English(EN) Task Vectors Can Be Linearly Decomposed into Interpretable and Controllable Components

新方法将大型语言模型任务向量分解为可解释的组件

作者 PulseAugur 编辑部 · [1 个来源] · 2026-11-02 05:00

研究人员开发了一种将“任务向量”（TVs）分解为可解释和可控组件的方法，打破了它们通常被视为整体方向的局限。这种新方法基于任务识别（TR）和任务学习（TL）原则，将每个TV分解为TR、TL和残差部分。通过贝叶斯优化，他们发现优化TR和TL系数比使用原始TV能更有效地进行干预，尤其是在基于指令的TV方面。这种分解提供了对TV如何运作的更清晰理解，并实现了对大型语言模型更精确的控制。 AI

影响这项研究可能带来更具可解释性和可控性的LLM引导机制，从而改善微调和模型适应性。

排序理由该集群包含一篇研究论文，详细介绍了一种分解LLM任务向量的新颖方法。[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 →

本文如何被排名

Signal score: 1 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇研究论文，详细介绍了一种分解LLM任务向量的新颖方法。[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: 1 days old; Coverage has settled into its steady-state source set.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.CL TIER_1 English(EN) · Yukiya Horiba, Hakaze Cho, Takahiro Yakoh · 2026-11-02 05:00

Task Vectors Can Be Linearly Decomposed into Interpretable and Controllable Components

arXiv:2610.41743v1 Announce Type: cross Abstract: Task Vectors (TVs) provide a compact way to steer large language models, but they are typically treated as atomized directions merely extracted from hidden states, leaving their internal functional structure poorly understood. Mot…

本文如何被排名

报道来源 [1]

Task Vectors Can Be Linearly Decomposed into Interpretable and Controllable Components

相关实体

相关话题