English

On Transferring Transferability: Towards a Theory for Size Generalization

Machine Learning 2026-02-12 v3 Representation Theory Statistics Theory Machine Learning Statistics Theory

Abstract

Many modern learning tasks require models that can take inputs of varying sizes. Consequently, dimension-independent architectures have been proposed for domains where the inputs are graphs, sets, and point clouds. Recent work on graph neural networks has explored whether a model trained on low-dimensional data can transfer its performance to higher-dimensional inputs. We extend this body of work by introducing a general framework for transferability across dimensions. We show that transferability corresponds precisely to continuity in a limit space formed by identifying small problem instances with equivalent large ones. This identification is driven by the data and the learning task. We instantiate our framework on existing architectures, and implement the necessary changes to ensure their transferability. Finally, we provide design principles for designing new transferable models. Numerical experiments support our findings.

Keywords

Cite

@article{arxiv.2505.23599,
  title  = {On Transferring Transferability: Towards a Theory for Size Generalization},
  author = {Eitan Levin and Yuxin Ma and Mateo Díaz and Soledad Villar},
  journal= {arXiv preprint arXiv:2505.23599},
  year   = {2026}
}

Comments

75 pages, 10 figures, closest to version to be published in NeurIPS

R2 v1 2026-07-01T02:48:42.348Z