Conference Tools for 2026 TMS Annual Meeting & Exhibition

Login

Register as a New User

Help

Submit An Abstract

Propose A Symposium

Presenter/Author Tools

Organizer/Editor Tools

About this Abstract

Meeting

2026 TMS Annual Meeting & Exhibition

Symposium

AI/ML/Data Informatics for Materials Discovery: Bridging Experiment, Theory, and Modeling

Presentation Title

Dis-GNN: Predicting Crystallographic Disorder Via Crystal Graph Neural Networks

Author(s)

Sam Dong, Ajinkya Hire, Jason Gibson, Richard Hennig

On-Site Speaker (Planned)

Sam Dong

Abstract Scope

The inverse design of functional materials is an unsolved problem that has the potential to evoke massive technological advances. Over the past few years, the accelerated progression of high fidelity machine learning models has led to a dramatic rise in efficiency in the generation/screening of inorganic crystals with targeted properties. A gap still remains between model predictions and what is observed through experiment. Disorder contributes to this gap. We introduce dis-GNN, a crystal-graph convolution neural network for disordered crystal classification and site occupancy prediction, achieving a hold-out accuracy of 95% on graph-level classification of disordered crystals, and a MAE of 0.05% for node-level predictions of site occupancies. From this, we survey the current landscape of generative machine learning for materials generation, evaluating the rate at which disordered crystals are predicted, highlighting the work that still needs to be done in the generative space for inverse materials design.

Proceedings Inclusion?

Planned:

Keywords

Machine Learning, Other,

OTHER PAPERS PLANNED FOR THIS SYMPOSIUM

A Computational Framework for the Intelligent Discovery of Selective Mineral Processing Reagents: Case Study of Phosphate Flotation

A Design-Focused Machine Learning Framework for Creep Behavior in Structural Alloys

A Multi-Objective Bayesian Optimization Framework for Predicting Grain-Boundary Segregation in High-Entropy Alloys

A Novel Approach for Predicting Mechanical Properties of Weld Metals Using a Physics-Informed Neural Network With Reduced Data Dependency Under Small-Data Regime

A PSO-ELM Based Prediction Model for Sulfur Content at the Endpoint of Hot Metal Pretreatment

A Transfer Learning Approach for Predicting Fatigue Behavior in Additively Manufactured Metals

A “DFT-ML” Framework for Rational Design of Next-Generation Semiconductors

Accelerating Materials Discovery with AI

Accelerating Materials Discovery with MaterialsFramework and PhaseForge: Universal Machine Learning Potentials and Automated Phase Diagram Generation

Accelerating Microstructural Solidification Simulations with Deep Neural Surrogates in Additive Manufacturing Conditions

AI-Driven Temperature Inference in XRD Patterns with Synthetic Pipelines and Molecular Dynamics Validation

An Agentic Framework for Extracting Structured Knowledge from Materials Science Literature

An Explainable Artificial Intelligence Strategy for Materials Design

Autonomous Control of an X-Ray Microscope: A Multi-Agent Framework for Closed-Loop Materials Characterization

Autonomous Exploration of Nanostructure Evolution in Solid-State Metal Dealloying via Laser-Induced Thermal Gradients and Multimodal Synchrotron X-Rays

Bayesian Inference of Local SiC/SiC Thermal Conductivity with Curved Geometry

Bayesian Inference of Local, Temperature-Dependent, Anisotropic Thermal Conductivity from Modulated Photothermal Phase Data

Bayesian Optimization for Inverse Design of Pt-Au Films with Custom Friction Properties

Beyond Atom-Centric: A Transformer-Based Approach for Crystal Structure Prediction

Beyond Gaussian Noise: Symmetric and Asymmetric Likelihood Models for Robust Materials Optimization

Can Machine Learning Predict Complex Materials and Materials Science Problems?

Comparing Universal Machine Learning Interatomic Potentials (uMLIP) in Calculating Selected Properties in Co–Ni–Ru Family Alloys

Creating Benchmark Datasets for Electron-Microscopy Based Machine Learning Models

Data-Driven and LLM-Based Design of Hydrogen Solubilities in Metallic Alloys

Decoding Non-Linearity and Complexity: Deep Tabular Learning Approaches for Materials Science

Deep Gaussian Process-Based Cost-Aware Batch Bayesian Optimization for Complex Materials Discovery Campaigns

Designing Novel High Entropy Alloys with High Strength and Ductility Using Machine Learning and Multi-Objective Optimization

Dis-GNN: Predicting Crystallographic Disorder Via Crystal Graph Neural Networks

Effects of Void Size and Material Properties on Hydrostatic Compression of a Single Crystal: A Combined Atomistic Simulation and Deep Learning Study

Evaluation of Out-of-Distribution Errors in Equivariant Graph Neural Networks for Materials Discovery

Extending First-Principles Simulations of Electronic Stopping Across Time and Length Scales Via Machine Learning

Foundations of Autonomy: AI-Ready Data Infrastructure for Labs and Digital Twins

From Atoms to Application: AI-Accelerated Materials Discovery in Constrained Design and Manufacturing Spaces

H-1: Artificial Intelligence in Water and Wastewater Treatment: A Meta-Analysis of Emerging Contaminants, Matrix Interference, and Real-Time Quality Monitoring

H-2: Enhancing the Hybrid Learning Experience: Navigating Lecture Videos with AI

H-3: Introducing XRDReader for Automated Extraction and Library Generation of X‑Ray Diffraction Data from Scientific Literature

H-5: Machine-Learned Spin-Lattice Dynamic Interatomic Potential for Iron-Manganese Alloys

H-7: Reconstruction of Polycrystalline Atomic Structure Using Diffusion Model

H-8: Transformer-Based Prediction of Mechanical Properties in CoCrCuFeNi High-Entropy Alloys

H-9: Universal Machine Learning Interatomic Potentials for Large-Scale Molecular Dynamics: Phase Transitions in Barium Titanate and Grain Boundary Segregation, Melting Behavior in High-Entropy Ceramics

Harmonized Data Schema for AI Ready Materials R&D Data

High-Throughput Study of Amorphous Dielectric Materials: Integrating Foundation Potentials with Density Functional Theory

How Do We Know When a New Materials Discovered? The "What is a Material Problem"

Influence of Microstructure on Mechanical Properties of High Entropy Alloys: A Physics-Informed Machine Learning Approach

Interpretable and Generative Deep Learning Framework for Predicting Flow Behavior and Microstructure Evolution in Thermo-Mechanical Processing

Large Language Model-Driven FAIR Materials Database for AI-Augmented Scientific Inquiry

Leveraging Active Learning and LLMs for Efficient Cold Spray Meta-Analysis

Leveraging Large Language Models to Optimize Materials Synthesis and Design

Machine Learning-Driven Exploration of Temperature and Pressure Dependent Phase Transitions in Ferroelectric Hafnium Dioxide

Machine Learning for Automated Decoding of Crystals from XRD Patterns

Machine Learning Framework for Predicting High-Strain-Rate Response Under Extreme Loading

Machine Learning Model to Predict Mass Loss During the Catalytic Pyrolysis of Polyetherimide/Graphite Nanocomposites

Modeling History Effects in Materials via a Recursive Neural Network

Modeling Stochastic Dynamics by Transforming Conditional Densities with Amortized Conditional Optimal Transport

Multimodal Spatial Encoding of Metal Microstructure and Plasticity for Mechanical Properties Prediction

Physics-Constrained Convolutional Neural Networks with Gradient-Aware Optimization for Real-Time Thermal Prediction in PBF Metal AM

Physics-Informed Neural Networks for Parameter Quantification in Materials Science

Posing Inverse Design of Mg-Alloy Microstructure and Texture as a Physics Guided Latent Diffusion-Optimization Problem

Predicting Dislocation Density Evolution via Machine Learned Symbolic Regression

Predicting Microstructural Outcomes in Heat-Treated Steels Using Machine Learning

Predicting the High-Temperature Oxidation Response of Nickel Superalloys Using CALPHAD-Enhanced Machine Learning

Prediction of Formation Energies of XYB₁₄ Boron-Rich Borides by Machine Learning

Prediction of Metal-Insulator-Transition (MIT) Compounds from Density of States (DOS) Data Using Image-Based Deep Learning

Quantum-Enhanced Machine Learning for HEA Discovery

Reconstructing Modal Shapes in Resonant Ultrasound Spectroscopy from Sparse Spatial Data via Attention-Based Machine Learning

Reinforcement Learning Recommendations for Crystallization-Resistant Nuclear Waste Glass Formulation

Simultaneous Microstructure and Composition-Based Machine Learning for Hardness Distribution Prediction & Verification in an Aluminum Alloy

Spin-Lattice Dynamics Study of Grain Boundary and Phase Interface Migration in Pure Iron

Streamlining Bayesian Optimization in Materials Science via a Retrieval-Augmented LLM Assistant Integrated with Honegumi

Text-to-Crystal Structure Generation Enabled by Fine-Tuned Large Language Model

Using 2D Data and Diffusion Model Principles to Generate 3D Microstructures of TRISO Fuel Compacts

Questions about ProgramMaster? Contact programming@programmaster.org | TMS Privacy Policy | Accessibility Statement