Distributed Representations of Atoms and Materials for Machine Learning

TL;DR

This paper introduces a method to create machine learning representations of atoms and compounds using only chemical formulas, achieving competitive results across various materials science tasks and proposing a new atom representation approach called SkipAtom.

Contribution

The paper presents a novel approach to derive compound representations from atomic embeddings and introduces SkipAtom for learning atom representations from structural data.

Findings

Compound representations are competitive with structure-based benchmarks.

Atom representations outperform existing methods in composition-only tasks.

SkipAtom effectively leverages structural database information.

Abstract

The use of machine learning is becoming increasingly common in computational materials science. To build effective models of the chemistry of materials, useful machine-based representations of atoms and their compounds are required. We derive distributed representations of compounds from their chemical formulas only, via pooling operations of distributed representations of atoms. These compound representations are evaluated on ten different tasks, such as the prediction of formation energy and band gap, and are found to be competitive with existing benchmarks that make use of structure, and even superior in cases where only composition is available. Finally, we introduce a new approach for learning distributed representations of atoms, named SkipAtom, which makes use of the growing information in materials structure databases.