Local Intrinsic Dimension of Representations Predicts Alignment and Generalization in AI Models and Human Brain

2026-01-30

AI Summary

This study investigates the shared mechanisms underlying the generalization capability of artificial intelligence models, inter-model alignment, and alignment between model representations and human brain neural activity. By introducing local intrinsic dimensionality as a unified geometric metric, the work overcomes the limitations of traditional global dimensionality approaches and demonstrates for the first time that this metric simultaneously predicts model generalization performance, cross-model consistency, and alignment with human brain representations. Integrating representational geometry analysis, local intrinsic dimensionality estimation, and large-scale comparisons between model representations and brain imaging data, the study finds that lower local intrinsic dimensionality correlates with stronger generalization and higher alignment across models and with the brain. Furthermore, increasing model scale and data size enhances performance by effectively reducing this dimensionality.

Technology Category

• Machine Learning: Representation Learning
• Humans and AI: Brain-Sensing and Analysis
• Natural Language Processing: (Large) Language Models

Application Category

• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Recent work has found that neural networks with stronger generalization tend to exhibit higher representational alignment with one another across architectures and training paradigms. In this work, we show that models with stronger generalization also align more strongly with human neural activity. Moreover, generalization performance, model--model alignment, and model--brain alignment are all significantly correlated with each other. We further show that these relationships can be explained by a single geometric property of learned representations: the local intrinsic dimension of embeddings. Lower local dimension is consistently associated with stronger model--model alignment, stronger model--brain alignment, and better generalization, whereas global dimension measures fail to capture these effects. Finally, we find that increasing model capacity and training data scale systematically reduces local intrinsic dimension, providing a geometric account of the benefits of scaling. Together, our results identify local intrinsic dimension as a unifying descriptor of representational convergence in artificial and biological systems.

Problem

• local intrinsic dimension
• representational alignment
• generalization
• model-brain alignment
• neural representations

Innovation

• local intrinsic dimension
• representational alignment
• generalization
• model-brain alignment
• scaling laws