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

2026-03-03

AI Summary

This work investigates how large language models adapt their internal representations in response to increasing task difficulty under out-of-distribution (OOD) inputs. The study reveals that as OOD shifts intensify—manifested through more complex reasoning, longer contexts, or a greater number of answer choices—the representations in the model’s final hidden layer become markedly sparser. Building on this observation, the authors propose Sparsity-Guided In-Context Learning (SG-ICL), a curriculum-based strategy that dynamically schedules demonstration examples according to the sparsity of their representations. Extensive experiments across diverse models and tasks demonstrate a consistent correlation between representation sparsity and OOD difficulty, and show that SG-ICL effectively enhances model performance in challenging OOD scenarios.

Technology Category

• Machine Learning: Representation Learning
• Search and Optimization: Learning to Search
• Natural Language Processing: (Large) Language Models

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

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.

Problem

• out-of-distribution
• representation sparsity
• large language models
• task difficulty
• hidden states

Innovation

• sparsity
• out-of-distribution
• large language models
• in-context learning
• representation dynamics