Presentation Information
[1P19]Unsupervised classification of boron nitride structures via UMAP of X-ray absorption spectra
*Reika Hasegawa1, Arpita Varadwaj1, Alexandre Lira Foggiatto1, Masahito Niibe2, Takahiro Yamazaki1, Masafumi Horio2, Yasunobu Ando3, Takahiro Kondo4, Iwao Matsuda2, Masato Kotsugi1 (1. Tokyo University of Science, 2. The University of Tokyo, 3. Institute of Science Tokyo, 4. University of Tsukuba)
X-ray absorption spectroscopy (XAS) provides valuable information on the structure and electronic states of materials. Its spectral profiles are complex, influenced by crystal structure and defects, and require extensive expertise for conventional analysis. This study presents an automated XAS analysis approach using various dimension reduction methods to characterize crystal and electronic structure. We show that, Uniform Manifold Approximation and Projection (UMAP), outperforms traditional methods in capturing the key features of complex spectra. Applying UMAP to simulated XAS spectra of boron nitride (BN), high-dimensional data can be precisely classified by crystal system, defect type, and layer number. By correlating UMAP-derived embeddings with electronic states, we identify nontrivial electronic properties. The successful application of this model to experimental data demonstrates its potential for autonomous structural identification, offering new possibilities for data-driven materials design and advancing novel material development.