Guardado en:
Detalles Bibliográficos
Autores principales: Yao, Junjie, Xu, Zhi-Qin John
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:https://arxiv.org/abs/2509.20124
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866909804666028032
author Yao, Junjie
Xu, Zhi-Qin John
author_facet Yao, Junjie
Xu, Zhi-Qin John
contents The embedding space of language models is widely believed to capture the semantic relationships; for instance, embeddings of digits often exhibit an ordered structure that corresponds to their natural sequence. However, the mechanisms driving the formation of such structures remain poorly understood. In this work, we interpret the embedding structures via the data distribution. We propose a set of probability signatures that reflect the semantic relationships among tokens. Through experiments on the composite addition tasks using the linear model and feedforward network, combined with theoretical analysis of gradient flow dynamics, we reveal that these probability signatures significantly influence the embedding structures. We further generalize our analysis to large language models (LLMs) by training the Qwen2.5 architecture on the subsets of the Pile corpus. Our results show that the probability signatures are faithfully aligned with the embedding structures, particularly in capturing strong pairwise similarities among embeddings. Our work uncovers the mechanism of how data distribution guides the formation of embedding structures, establishing a novel understanding of the relationship between embedding organization and semantic patterns.
format Preprint
id arxiv_https___arxiv_org_abs_2509_20124
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Probability Signature: Bridging Data Semantics and Embedding Structure in Language Models
Yao, Junjie
Xu, Zhi-Qin John
Machine Learning
The embedding space of language models is widely believed to capture the semantic relationships; for instance, embeddings of digits often exhibit an ordered structure that corresponds to their natural sequence. However, the mechanisms driving the formation of such structures remain poorly understood. In this work, we interpret the embedding structures via the data distribution. We propose a set of probability signatures that reflect the semantic relationships among tokens. Through experiments on the composite addition tasks using the linear model and feedforward network, combined with theoretical analysis of gradient flow dynamics, we reveal that these probability signatures significantly influence the embedding structures. We further generalize our analysis to large language models (LLMs) by training the Qwen2.5 architecture on the subsets of the Pile corpus. Our results show that the probability signatures are faithfully aligned with the embedding structures, particularly in capturing strong pairwise similarities among embeddings. Our work uncovers the mechanism of how data distribution guides the formation of embedding structures, establishing a novel understanding of the relationship between embedding organization and semantic patterns.
title Probability Signature: Bridging Data Semantics and Embedding Structure in Language Models
topic Machine Learning
url https://arxiv.org/abs/2509.20124