论文概要
研究领域: CV 作者: Aaron Sun, Yachan Liu, Ping Yang, Peng Bai, Subhransu Maji 发布时间: 2026-11-06 arXiv: 2611.09880
Overview
Zeolites are an important class of porous crystalline materials with broad applications in gas storage, separations, and catalysis. Although hundreds of thousands of experimentally realized and hypothetical zeolite structures are known, identifying materials optimized for a target application remains challenging because property labels are expensive to obtain and available only for limited structure-molecule pairs.
While prior work has demonstrated that 3D neural representations can predict zeolite properties effectively, it remains unclear which representation-learning strategies are most effective in the low-data regime relevant to materials discovery.
Key Points
- ZeoBench is a benchmark for self-supervised learning on dense 3D representations of all-silica zeolites.
- Learned features are evaluated on six existing downstream adsorption-property prediction tasks across a range of training-set sizes.
- Strategies compared include:
- Self-supervised 3D pretraining on zeolite volumetric data
- Adapting image-pretrained 2D ConvNets and ViTs to 3D via multi-view and channel-inflation strategies
- 3D ConvNets pretrained on out-of-domain volumetric datasets
- Handcrafted descriptors and supervised baselines
- Findings: Self-supervised 3D representations outperform existing methods, achieving state-of-the-art on both in-domain label regimes and the unseen metal-organic framework domain. Learned representations consistently beat handcrafted descriptors, and convolutional architectures generally prove more effective than vision Transformers in this setting.