2025 · AI for Accelerated Materials Design Workshop, ICLR 2025
TDCM25: A multi-modal, multi-task benchmark for temperature-dependent crystalline materials
Can Polat , Hasan Kurban , Erchin Serpedin , Mustafa KurbanResearch highlights
TDCM25: one material across temperatures and phases
Benchmark temperature-dependent crystalline materials through paired structural, visual, and textual inputs.
Why we investigated it
Temperature and phase change the properties a model must predict, even when the underlying chemistry stays fixed.
What the work contributes
TDCM25 organizes TiO2 examples across anatase, brookite, and rutile with multiple temperatures and modalities.
What the evidence shows
The benchmark contains approximately 100,000 entries across 21 temperatures, enabling controlled comparisons of phase- and temperature-sensitive tasks.
Build on this work
Use the benchmark to design phase classification, property prediction, and explanation tasks with explicit temperature controls.
- Temperature-dependent materials
- TiO2
- Multimodal benchmarks
- Phase classification
Cite
@article{polat2025tdcm25,
title = {TDCM25: A multi-modal, multi-task benchmark for temperature-dependent crystalline materials},
author = {Can Polat and Hasan Kurban and Erchin Serpedin and Mustafa Kurban},
year = 2025,
journal = {AI for Accelerated Materials Design Workshop, ICLR 2025},
url = {https://openreview.net/pdf?id=bNB5SQTqKL},
}