Graph Neural Network Expressivity and Meta-Learning for Molecular Property Regression

Haitz Sáez de Ocáriz Borde; Federico Barbero

Graph Neural Network Expressivity and Meta-Learning for Molecular Property Regression

Machine Learning 2022-11-28 v2 Biomolecules Quantitative Methods

Authors: Haitz Sáez de Ocáriz Borde , Federico Barbero

Abstract

We demonstrate the applicability of model-agnostic algorithms for meta-learning, specifically Reptile, to GNN models in molecular regression tasks. Using meta-learning we are able to learn new chemical prediction tasks with only a few model updates, as compared to using randomly initialized GNNs which require learning each regression task from scratch. We experimentally show that GNN layer expressivity is correlated to improved meta-learning. Additionally, we also experiment with GNN emsembles which yield best performance and rapid convergence for k-shot learning.

Keywords

graph neural network extreme learning machine neural network

Cite

@article{arxiv.2209.13410,
  title  = {Graph Neural Network Expressivity and Meta-Learning for Molecular Property Regression},
  author = {Haitz Sáez de Ocáriz Borde and Federico Barbero},
  journal= {arXiv preprint arXiv:2209.13410},
  year   = {2022}
}

Related papers

View all related →