Local Intrinsic Dimension of Representations Predicts Alignment and Generalization in AI Models and Human Brain
This work provides a geometric explanation for generalization and alignment in AI and neuroscience, offering insights into scaling effects, but it is incremental as it builds on prior findings about representational alignment.
The paper tackled the problem of understanding generalization and alignment in AI models and the human brain, showing that lower local intrinsic dimension of representations predicts stronger model-model and model-brain alignment and better generalization, with systematic reductions in local dimension observed with increased model capacity and data scale.
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.