跳到正文
arXiv cs.CL· Francesco Cagnetta, Santiago Acevedo·· 2 天前AI 评分31

深层隐变量、长记忆:隐变量持续性解释 Transformer 激活中的时间相关性

Deep Latents, Long Memory: Latent persistence explains temporal correlations in transformer activations

AI 导读

研究提出用层激活的自相关函数刻画 LLM 隐层表征,发现自相关衰减尾部随层数加深而变长,在合成与真实数据中均成立。以稀疏自编码器特征作为隐变量代理时,特征持续性随深度增加、与自相关强度高度相关,且消融这些持续性特征后自相关随之下降。

正文 · 原文

View PDF HTML (experimental)

Abstract:The hidden layers of Large Language Models (LLMs) are believed to encode representations of the data whose abstraction increases with depth. However, in the absence of an operational notion of abstraction, this statement is hard to test quantitatively. Here, we propose an alternative view, building on the principle that the correlations of a sequence reflect the structure of the process that generated it. We define the autocorrelation function of a layer as the average dot product between its activations at two context positions separated by a lag. Latent variables encoded persistently along the context create slowly decaying tails in the autocorrelation, and we find these tails to grow across layers in both synthetic and natural data. In synthetic data with controlled hierarchical latent structure, this growth stems from the encoding of abstract latent variables, which govern long stretches of the input. Indeed, the persistence of latent codes predicts the qualitative structure of autocorrelations and accounts for most of the empirical measures. In real text, we take sparse-autoencoder features as proxies for latent variables, and find that their persistence correlates strongly with the magnitude of autocorrelations across datasets and LLMs. In particular, the persistence time increases with depth, as do autocorrelations, implying that deeper latent features persist for longer context spans. We complement this analysis by explicitly ablating the persistent sparse-autoencoder features and observing a consequent decrease in the autocorrelations.
Subjects: Machine Learning (cs.LG); Computation and Language (cs.CL)
Cite as: arXiv:2611.18089 [cs.LG]
  (or arXiv:2611.18089v1 [cs.LG] for this version)
  https://doi.org/10.48550/arXiv.2611.18089

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Francesco Cagnetta [view email]
[v1] Fri, 13 Nov 2026 12:14:17 UTC (4,238 KB)

来源:arXiv cs.CL · arxiv.org