arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2603.06592cs.CLcs.LG
收藏

数据生成过程中层次化潜在结构在跨尺度的机理现象中统一

Hierarchical Latent Structures in Data Generation Process Unify Mechanistic Phenomena across Scale

Ubiquitous Knowledge Processing Lab (UKP Lab) Department of Computer Science, Technical University of Darmstadt and National Research Center for Applied Cybersecurity ATHENE, Germany(通用知识处理实验室(UKP Lab)计算机科学系,达姆施塔特技术大学和应用网络安全国家研究中心ATHENE,德国)

Jonas Rohweder, Subhabrata Dutta, Iryna Gurevych

AI总结:

本文通过生成合成语料库研究数据生成过程中层次结构对语言模型中归纳头、功能向量和海德拉效应等现象的统一解释。

摘要:

当代研究揭示了基于Transformer的语言模型神经信息处理中许多令人困惑的现象。建立这些现象的稳健、统一理解需要在模型训练范围内拆解模型。虽然预训练语料库的不可捉摸规模限制了在此方向上的自底向上研究,但对数据生成过程的简单假设限制了表达力,并未能解释复杂模式。在本工作中,我们使用概率上下文无关文法(PCFGs)生成合成语料库,这些语料库是忠实且计算高效的网络规模文本语料库的代理。我们调查了在我们设计的数据生成过程中,三个机理现象——归纳头、功能向量和海德拉效应的出现情况,以及在真实世界语言模型的检查点中。我们的发现表明,数据生成过程中的层次结构在解释这些现象的出现中起到了决定性作用。我们提供了层次在语言模型训练动态中所起作用的理论基础。简而言之,我们的工作是首次为LLMs中看似不相关的机理现象的出现提供统一解释,并辅以高效的合成工具用于未来可解释性研究。

英文摘要:

Contemporary studies in mechanistic interpretability have uncovered many puzzling phenomena in the neural information processing of Transformer-based language models, such as induction heads, function vectors, and the Hydra effect. Some of these individual phenomena have been independently tied to different data distributional properties, while some have been loosely associated with model architecture and how Transformers process information. However, a unified understanding of the relationship between data, model architecture, and optimization remains lacking, failing to answer the fundamental question: why do these three phenomena appear universally across different model families and scales, despite their seeming disconnect? In this work, we answer this question by unifying these three phenomena as consequences of hierarchical latent structures in the data generation process, coupled with decorrelated gradients across additive model components and directional concavity in the representation geometry. We validate our theoretical results in a toy model regime and in a large-scale synthetic data regime, comparing them with language models trained on natural language data.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑