Farther the Shift, Sparser the Representation: Analyzing OOD Mechanisms in LLMs

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Jin, Mingyu, Yin, Yutong, Niu, Jingcheng, Zeng, Qingcheng, Xu, Wujiang, Du, Mengnan, Cheng, Wei, Wang, Zhaoran, Chen, Tianlong, Metaxas, Dimitris N.
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866914409122627584
author Jin, Mingyu
Yin, Yutong
Niu, Jingcheng
Zeng, Qingcheng
Xu, Wujiang
Du, Mengnan
Cheng, Wei
Wang, Zhaoran
Chen, Tianlong
Metaxas, Dimitris N.
author_facet Jin, Mingyu
Yin, Yutong
Niu, Jingcheng
Zeng, Qingcheng
Xu, Wujiang
Du, Mengnan
Cheng, Wei
Wang, Zhaoran
Chen, Tianlong
Metaxas, Dimitris N.
contents In this work, we investigate how Large Language Models (LLMs) adapt their internal representations when encountering inputs of increasing difficulty, quantified as the degree of out-of-distribution (OOD) shift. We reveal a consistent and quantifiable phenomenon: as task difficulty increases, whether through harder reasoning questions, longer contexts, or adding answer choices, the last hidden states of LLMs become substantially sparser. In short, \textbf{\textit{the farther the shift, the sparser the representations}}. This sparsity--difficulty relation is observable across diverse models and domains, suggesting that language models respond to unfamiliar or complex inputs by concentrating computation into specialized subspaces in the last hidden state. Through a series of controlled analyses with a learning dynamic explanation, we demonstrate that this sparsity is not incidental but an adaptive mechanism for stabilizing reasoning under OOD. Leveraging this insight, we design \textit{Sparsity-Guided Curriculum In-Context Learning (SG-ICL)}, a strategy that explicitly uses representation sparsity to schedule few-shot demonstrations, leading to considerable performance enhancements. Our study provides new mechanistic insights into how LLMs internalize OOD challenges. The source code is available at the URL: https://github.com/MingyuJ666/sparsityLLM.
format Preprint
id arxiv_https___arxiv_org_abs_2603_03415
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Farther the Shift, Sparser the Representation: Analyzing OOD Mechanisms in LLMs
Jin, Mingyu
Yin, Yutong
Niu, Jingcheng
Zeng, Qingcheng
Xu, Wujiang
Du, Mengnan
Cheng, Wei
Wang, Zhaoran
Chen, Tianlong
Metaxas, Dimitris N.
Computation and Language
Artificial Intelligence
In this work, we investigate how Large Language Models (LLMs) adapt their internal representations when encountering inputs of increasing difficulty, quantified as the degree of out-of-distribution (OOD) shift. We reveal a consistent and quantifiable phenomenon: as task difficulty increases, whether through harder reasoning questions, longer contexts, or adding answer choices, the last hidden states of LLMs become substantially sparser. In short, \textbf{\textit{the farther the shift, the sparser the representations}}. This sparsity--difficulty relation is observable across diverse models and domains, suggesting that language models respond to unfamiliar or complex inputs by concentrating computation into specialized subspaces in the last hidden state. Through a series of controlled analyses with a learning dynamic explanation, we demonstrate that this sparsity is not incidental but an adaptive mechanism for stabilizing reasoning under OOD. Leveraging this insight, we design \textit{Sparsity-Guided Curriculum In-Context Learning (SG-ICL)}, a strategy that explicitly uses representation sparsity to schedule few-shot demonstrations, leading to considerable performance enhancements. Our study provides new mechanistic insights into how LLMs internalize OOD challenges. The source code is available at the URL: https://github.com/MingyuJ666/sparsityLLM.
title Farther the Shift, Sparser the Representation: Analyzing OOD Mechanisms in LLMs
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2603.03415