Conference Tools for 2027 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

2027 TMS Annual Meeting & Exhibition

Symposium

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

Presentation Title

Data-efficient Deep Generative Modeling for Microstructure Prediction under Unexplored Compositional Conditions

Author(s)

Xiaofan Zhang, Junya Inoue, Satoshi Noguchi

On-Site Speaker (Planned)

Xiaofan Zhang

Abstract Scope

Microstructure prediction under limited experimental data remains a significant challenge in AI-assisted materials discovery. This study presents a data-efficient deep generative framework that integrates a Vector-Quantized Variational Autoencoder (VQVAE) with a conditional Pixel Convolutional Neural Network (PixelCNN) to learn compact latent representations of complex microstructures and generate realistic microstructures conditioned on material descriptors, including composition, processing parameters, and other measurable characteristics. Whereas previous studies have developed and evaluated generative models using microstructures from all available material datasets, the present study investigates whether realistic microstructures can be predicted for material conditions whose corresponding microstructures are unavailable to the model during training, based solely on the associated material descriptors. This setting reflects a practical challenge in materials research, where experimental observations are often unavailable for newly explored material conditions. The proposed framework aims to extend the applicability of data-efficient generative modeling to microstructure prediction beyond available experimental datasets, supporting AI-assisted materials discovery.

Proceedings Inclusion?

Planned:

Keywords

Machine Learning, Computational Materials Science & Engineering, Modeling and Simulation

OTHER PAPERS PLANNED FOR THIS SYMPOSIUM

A Comparative Study of Ensemble Machine Learning Models for Nanoindentation Hardness and Elastic Modulus of Ion-irradiated HEAs

A High throughput experimental-computational approach linking Stacking Fault Energy and nanotwin formation in Nickel-based thin film libraries

A Multidimensional Continuous Conditional Generative Adversarial Network to Link Composition-Process-Microstructure-Properties in Mg Alloys

A Paired Benchmark of Tight-Binding and Graph Neural Networks for Electronic Bandgap Prediction

A Physics-Informed Inverse Design Pipeline for Halogen-Free Fire-Retardant Polymers

A Self-Driving Additive Manufacturing Platform for Autonomous Superalloy Design: A case study on Co-superalloys

Accelerating Structural Materials Discovery by Orders of Magnitude Using Random Libraries

Advancing BEE-NET for Physics-Informed Quantum Materials Discovery

Agentic AI-Driven Design of Refractory Multi-Principal Element Alloys

Agentic AI for Assessing and Optimizing Scientific Reasoning and Creativity in Chemical Vapor Deposition Diamond Synthesis

Agentic Lifestyles in the Era of AI: Research Workflows and Insights from the Vertical Cloud Lab @ BYU

Agentic Resource Allocation for Batch Multi-Objective Bayesian Optimization in Autonomous Materials Discovery

AI-driven theoretical frameworks to determine structure of complex materials and defects from experiments

AI-Enabled High Throughput Discovery of High Entropy Alloys for Extreme Environments

An Automated Probabilistic Machine Learning Framework for Materials Modeling and Active Learning: Benchmarking Uncertainty Quantification Across Analytical, Computational, and Experimental Domains

An informatics approach to predict shape memory effect in CrMnFeCoNi high-entropy alloys

Any Data is AI-Ready via Semantic Knowledge Graphs

Areas of Agentic Leverage in Computational Materials Science

Assessing Uncertainty and Reproducibility in ML-Based Grain Growth Predictions: Impact of Training Data

Automating LabDCT Workflows via Intelligent, Data-Driven Guidance

CALIBER: A Retrieval-Augmented, Uncertainty-Aware Platform for Selecting High-Fidelity Characterization Parameters for Novel Alloys

Causally Informed Machine Learning for Nanoporous Microstructure-Property Relationships

COD'HEM: An Integrated Experimental–DFT Materials Informatics Platform for Data-Driven Discovery of High Entropy Materials

Damage assessment digital twin for fusion plasma-facing materials

Data-driven inference of twin network dynamics: nucleation, co-nucleation, transmission, and incidental contacts

Data-Driven Surrogate to FE for Predicting Elastoplastic Constitutive Relation in Polymer Nanocomposites

Data-Driven Tensile Strength Prediction of Biodegradable Zinc Alloys

Data-efficient Deep Generative Modeling for Microstructure Prediction under Unexplored Compositional Conditions

Deep learning based optimization of hydrogels and their funcationalization for applications in ophthalmology

Design of Experiments for Efficient XRF Under Compositional Uncertainty

Developing Methods, Models, and Datasets for a Self-Driving Magnetic Nanomaterials Laboratory: Application for Thermal Magnetic Particle Imaging

Development of Machine-Learning Interatomic Potential forSolute–Dislocation Interactions in Fe–C–Cu Steels

Dis-GNN: Crystal Graph Neural Network for Disordered Crystal Classification and Site Occupancy Prediction

Dual Machine Learning Discovery of the Radius of Informative Structural Environments in Metallic Glasses

Fine–Tuned Universal Machine Learned Interatomic Potentials for Surface Property Prediction

From Loading to Architecture: A Physics-Guided Machine-Learning Framework for Polymer Spall Strength

Graph Neural Network Optimization of Industrial Materials, Formulations, and Chemical Processes

Graph Neural Network Prediction of Grain-Boundary Diffusion from a Reduced Active-Atom Subset

Graph Neural Networks for the Prediction of Metastable Grain Boundary Dynamics

Hyperparameter-Informed Sparse Bayesian Active Learning for Understanding and Controlling Frontal Radical Polymerization

Integrating AI-enabled analysis and agentic workflows at scientific user facilities

Interpretable Machine Learning and Topological Descriptors for Structural Evolution, Interfaces, and Dopant Thermodynamics: A Halide Perovskites Case Study

Interrogating Materials Informatics Models Along Experimentally Relevant Alloying Pathways

Knowledge-Evolution-Informed Forecasting of Mg Alloy Research Using Large Language Models

Label-Efficient Strategies for Small-Data Materials Modeling with a Nuclear Waste Glass Case Study Study

Language-Based Modeling for Additive-Manufacturing Alloy Design: Learning, Reasoning, and Generative Exploration

Learning Across Theories: A Multi-Fidelity Machine-Learned Interatomic Potential for the Energetic Molecular Crystal LLM-105

Learning topological evolution in polycrystalline microstructures

Machine-Learned Interatomic Potentials for High-Throughput Screening of Ag-Exchanged Zeolites for Xenon Capture in Nuclear Non-Proliferation Sensing

Machine learning and high-throughput computational guided development of high temperature oxidation-resisting Ni-Co-Cr-Al-Fe based high-entropy alloys

Machine Learning Framework for Predicting and Interpreting Corrosion Behavior of Zinc Alloys

Machine Learning Interatomic Potential for Titanium-Helium systems

Machine learning to derive effective transport properties of multiphase polycrystalline systems

Mag-GPT: From Query to Magnetic Materials Discovery

Million-scale Surface Property Screening Enabled by Universal Machine Learning Interatomic Potentials

Multi-Scale Microstructure–Property Prediction Using Deep Learning

Multi-Surrogate Strategies for Multi-Objective Bayesian Optimization: A Cross-Phase Empirical Study from Static Baselines to LLM-Proposed Policies

Multiscale Modeling of GaN/AlGaN Interfaces with Machine Learning and DFT

Out-of-Distribution Phase Equilibrium Prediction Using Physics-Informed Graph Attention Networks

Physics-Informed Constitutive Calibration as a Data Problem: From Quasi-Static-to-High-Rate Experiments and FEA Validation

Physics-Informed Graph Neural Networks for Predicting Li-Ion Transport in Solid Electrolyte Interphases

Physics-informed Machine Learning Framework for Material Parameters Estimation from Materials Morphology

Physics-Informed Machine Learning Framework for Predicting Electromigration Failure in Solder Joints

Physics Informed Neural Network Recovery of Continuous Relaxation Time Spectra in Linear Viscoelasticity

Process-aware Extraction of Composition–processing–property Records from Materials Literature Using Large Language Models

Research on Molten Steel Carbon Content Prediction for Multi-Procedure Steelmaking Process

Resource-Constrained AI Discovery of Manufacturable Wear-Resistant Steels

Self-Supervised Learning of Materials States Evolution from Microscopy Data Using Joint Embedding Predictive Architectures

Steel S-N curve prediction via domain knowledge guided LLM-assisted symbolic regression

The Surface Problem in Machine-Learned Interatomic Potentials

Thermal properties dataset of high-emissivity ceramic materials generated with automated data extraction

Toward an Autonomous Laboratory for Accelerated Alloy Discovery

Toward Closed-Loop Materials Design: AI Prediction, Uncertainty, and Decision-Making

Trustworthy, Uncertainty-Aware Multimodal Machine Learning for Materials Discovery: Interpretable, Self-Curating Screening of Solid-State Electrolytes

Uncertainty-Guided Active Learning for Autonomous Spall Strength Experimentation

User-Friendly CALPHAD Workflows for Materials Discovery Beyond the CALPHAD Community

XRDReader: Turning Published XRD Information into Machine-Readable Experimental Databases with Multimodal Agentic AI

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