研究目的
Exploring the space of materials electronic band structures using unsupervised machine learning to analyze and visualize high-dimensional data.
研究成果
t-SNE was successfully applied to explore high-dimensional feature spaces defined by electronic band structure features, clustering materials with similar dispersion curves. The algorithm presents a powerful way to explore the band structure space, allowing for the visualization of thousands of materials and the exploration of similarities and differences on that feature space.
研究不足
The probabilistic nature of the t-SNE algorithm means the mapping is not unique, and the algorithm does not guarantee to correctly map large distances. The final mapping depends strongly on the parameters of the algorithm, particularly the Perplexity parameter.
1：Experimental Design and Method Selection:
The t-SNE algorithm was applied to analyze multidimensional spaces defined by features extracted from the electronic band structure of materials.
2：Sample Selection and Data Sources:
The dataset consisted of approximately 2500 metal compounds with crystalline structures belonging to the I mm4 symmetry group.
3：List of Experimental Equipment and Materials:
Band structures were calculated with Density functional theory (DFT) by AFLOW.
4：Experimental Procedures and Operational Workflow:
Feature vectors were defined by sampling Ek values from the electronic band structures, and t-SNE was used to map these high-dimensional spaces to lower dimensions.
5：Data Analysis Methods:
The KL divergence was minimized to optimize the positions in the low-dimensional space, and the influence of the Perplexity parameter on the mapping was analyzed.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
