Abstract:Crystal graph convolutional neural networks (CGCNNs) have revolutionized materials research by eliminating the need for manual feature engineering. However, their lack of interpretability and sensitivity to structural distortions hinders their application in substitution engineering. Therefore, we propose an innovative adversarial example method for guiding feature construction and enhancing the interpretability of CGCNNs. In this study, our focus lies on identifying CsSnBr 3− x I x configurations with high-energy conversion efficiency. Initially, we train a CGCNN classifier as a benchmark. Subsequently, we perturb input data to generate a low-performance classifier and identify adversarial examples based on incorrect predictions. Upon comparing these examples with normal examples, we observe substantial structural distortions in adversarial cases, serving as inspiration for the creation of disorder-related features. Consequently, an interpretable model is developed, which surpasses CGCNNs with atomic position perturbations and a gradient-boosting classifier using general features. Notably, the previously overlooked feature "number of unequal atoms" plays an important role in offering crucial insights. Further analysis reveals that configurations with pronounced disorder can exhibit increased power density, thereby enhancing the energy conversion efficiency. Our work not only elucidates the impact of atom substitution on energy conversion efficiency but also provides a roadmap for constructing interpretable machine learning models.

What problem does this paper attempt to address?

The paper primarily aims to address the following issues: ### Research Background and Objectives - **Challenges in Material Screening and Design**: In the exploration of high-performance perovskite materials, traditional trial-and-error methods (such as Density Functional Theory (DFT) calculations and experiments) are time-consuming and costly. - **Application of Machine Learning**: Machine learning techniques can accelerate the discovery process of new materials, especially by using methods like Crystal Graph Convolutional Neural Networks (CGCNNs) to analyze crystal structure properties. - **Limitations of CGCNNs**: Although CGCNNs perform well in material research, their lack of interpretability and sensitivity to structural distortions limit their application in alternative engineering. ### Main Issues Addressed - **Enhancing the Interpretability of CGCNNs**: Propose an adversarial sample-based approach to enhance the feature construction and model interpretability of CGCNNs. - **Identifying High Energy Conversion Efficiency CsSnBr3−xIx Configurations**: The goal of the study is to identify CsSnBr3−xIx configurations with high energy conversion efficiency, with a particular focus on the substitution of bromine (Br) with iodine (I). - **Developing Interpretable Machine Learning Models**: By improving CGCNNs and utilizing adversarial samples, develop a machine learning model that surpasses traditional CGCNNs and is easier to understand. ### Method Overview - **Training the CGCNN Baseline Model**: First, train a CGCNN classifier as a baseline. - **Generating Adversarial Samples**: Generate adversarial samples by slightly perturbing the input data to create a lower-performing classifier and identify adversarial samples based on incorrect predictions. - **Feature Construction**: Compare adversarial samples with normal samples, finding that adversarial samples exhibit greater structural distortions, which inspires the construction of disorder-related features. - **Developing an Interpretable Model**: Based on these features, develop an interpretable machine learning model that not only outperforms the CGCNNs model with atomic position perturbations but also surpasses the Gradient Boosting Classifier (GBC) using general features. ### Results and Contributions - **Key Feature**: The feature "number of unequal atoms" is crucial for providing important insights. - **Performance Improvement**: The proposed interpretable model excels in predicting the Spectroscopic Limited Maximum Efficiency (SLME) of CsSnBr3−xIx configurations. - **Mechanism Revelation**: The study shows that configurations with significant disorder can exhibit higher power density, thereby improving energy conversion efficiency. In summary, this research aims to optimize alternative engineering strategies to discover high-performance perovskite materials by enhancing the interpretability of machine learning models.

Enhancing interpretability in the exploration of high-energy conversion efficiency in CsSnBr3−xIx configurations using crystal graph convolutional neural networks and adversarial example methods

Developing an improved Crystal Graph Convolutional Neural Network framework for accelerated materials discovery

Study of crystal properties based on attention mechanism and crystal graph convolutional neural network

Crystal Graph Convolutional Neural Networks for an Accurate and Interpretable Prediction of Material Properties

Enhancing crystal structure prediction by combining computational and experimental data via graph networks

General Graph Neural Network-Based Model To Accurately Predict Cocrystal Density and Insight from Data Quality and Feature Representation

CEGANN: Crystal Edge Graph Attention Neural Network for multiscale classification of materials environment

Towards accurate prediction of configurational disorder properties in materials using graph neural networks

Effective optimization of atomic decoration in giant and superstructurally ordered crystals with machine learning

CTGNN: Crystal Transformer Graph Neural Network for Crystal Material Property Prediction

Graph Neural Networks for an Accurate and Interpretable Prediction of the Properties of Polycrystalline Materials

CrysToGraph: A Comprehensive Predictive Model for Crystal Materials Properties and the Benchmark

Towards invertible 2D crystal structure representation for efficient downstream task execution

CrysGraphFormer: an equivariant graph transformer for prediction of lattice thermal conductivity with interpretability

Tripartite interaction representation algorithm for crystal graph neural networks

A machine learning‐based crystal graph network and its application in development of functional materials

An effective method for generating crystal structures based on the variational autoencoder and the diffusion model

GPT-assisted learning of structure-property relationships by graph neural networks: Application to rare-earth doped phosphors

Learning Ordering in Crystalline Materials with Symmetry-Aware Graph Neural Networks

Coarse-grained crystal graph neural networks for reticular materials design