Reward driven discovery of the optimal microstructure representations with invariant variational autoencoders

Extracting physically meaningful information from microscopy images requires selecting suitable structural descriptors and constructing low-dimensional representations. However, the quality of these representations strongly depends on the choice of descriptors, model architecture, and hyperparameters, which are typically selected empirically. Variational Autoencoders (VAEs) provide a powerful means of constructing such low-dimensional representations, but their performance heavily depends on multiple non-myopic design choices, which are often optimized through trial-and-error and empirical analysis. To enable automated and unbiased optimization of VAE workflows, we investigated reward-based strategies for evaluating latent space representations. Using Piezoresponse Force Microscopy imaging as a model system, we examined multiple reward functions for automated VAE workflow optimization. Our analysis shows that Gaussian Mixture Models (GMM) and Bayesian Gaussian Mixture Models (BGMM) provide a basis for constructing reward functions that balance representation quality, complexity, and cluster separability. In the demonstrated case, the ICL-based reward identified four dominant components corresponding to distinct microstructural configurations, illustrating how reward-based optimization can guide the discovery of parsimonious representations.

Authors

Institutions

Publication Details

Journal
Nanotechnology
Published
2026-11-17
DOI
https://doi.org/10.1088/1361-6528/aec2fd
Citations
1
Primary Topic
Machine Learning in Materials Science
Type
article
Field-Weighted Citation Impact
0.87

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Reward driven discovery of the optimal microstructure representations with invariant variational autoencoders

Maxim A Ziatdinov, Hiroshi Funakubo, Doru Constantin Lupascu, Kamyar Barakati et al.
1 citations
Nanotechnology
Machine Learning in Materials Science
0.87
article

Reward driven discovery of the optimal microstructure representations with invariant variational autoencoders

Maxim A Ziatdinov, Hiroshi Funakubo, Doru Constantin Lupascu, Kamyar Barakati, Sergei V. Kalinin, Vladimir V. Shvartsman, Boris N. Slautin
article en
1 citations

Abstract

Extracting physically meaningful information from microscopy images requires selecting suitable structural descriptors and constructing low-dimensional representations. However, the quality of these representations strongly depends on the choice of descriptors, model architecture, and hyperparameters, which are typically selected empirically. Variational Autoencoders (VAEs) provide a powerful means of constructing such low-dimensional representations, but their performance heavily depends on multiple non-myopic design choices, which are often optimized through trial-and-error and empirical analysis. To enable automated and unbiased optimization of VAE workflows, we investigated reward-based strategies for evaluating latent space representations. Using Piezoresponse Force Microscopy imaging as a model system, we examined multiple reward functions for automated VAE workflow optimization. Our analysis shows that Gaussian Mixture Models (GMM) and Bayesian Gaussian Mixture Models (BGMM) provide a basis for constructing reward functions that balance representation quality, complexity, and cluster separability. In the demonstrated case, the ICL-based reward identified four dominant components corresponding to distinct microstructural configurations, illustrating how reward-based optimization can guide the discovery of parsimonious representations.

Nanotechnology
Tokyo Institute of Technology (JP), Oak Ridge National Laboratory (US), Pacific Northwest National Laboratory (US), Center for Nanophase Materials Sciences (US), University of Duisburg-Essen (DE), University of Tennessee at Knoxville (US)
National Science Foundation, U.S. Department of Energy, Battelle, Ministry of Education, Culture, Sports, Science and Technology, Materials Research Science and Engineering Center, Harvard University, Japan Science and Technology Agency, Pacific Northwest National Laboratory
Openalex Percentile: Top 36%
Machine Learning in Materials Science
0.87
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon