Training Data and Models for Predicting Rotation-Invariant Material Properties

A 186.3 MB collection of training datasets, code, and pre-trained convolutional neural network models for materials science research. The data supports the paper 'Convolutional neural networks for learning rotation-invariant properties of both amorphous and crystalline materials' published in Phys. Rev. Materials in 2026. The author is Zhao Fan, and the dataset was last updated on May 14, 2026.

Use Cases

• Training CNN models for predicting rotation-invariant properties of materials based on the provided training datasets.
• Computing structural descriptor matrices (SDMs) for materials using the included code.
• Applying pre-trained CNN models to predict properties of new amorphous or crystalline materials.
• Reproducing the results of the associated research paper on rotation-invariant material property prediction.

Strengths

• Includes well-trained CNN models, reducing the need for initial training.
• Provides both training data and the code for computing structural descriptor matrices.
• Dataset size is 186.3 MB, indicating a medium-scale resource.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale training.

Provenance

Source

Zhao Fan

Collection Method

Likely generated from computational materials science simulations or experiments.

Freshness

Last updated 2026-05-14 18:25:33