[14p-K404-1~17] 12.2 Characterization and Materials Physics

4:45 PM - 5:00 PM JST(7:45 AM - 8:00 AM UTC)

[14p-K404-13] Development of a clustering method for automatic classification of electrical conductance data from single molecule junctions measured via STM-BJ method

〇(B)Kyo Ogino¹, Ryo Yamada², Hirokazu Tada² (1.School of Engineering Science Osaka Univ., 2.Grad. School of Engineering Science Osaka Univ.)

Keywords:

single molecule electronics,machine learning

The electronic transport properties of single-molecule junctions have been extensively studied since the advent of the break junction (BJ) method. In the BJ method, it is necessary to classify and analyze an enormous dataset consisting of tens of thousands of measurements. While unsupervised machine learning (clustering) is useful for such data classification, it presents a challenge in that the results heavily depend on the choice of features. In this study, we developed a general clustering approach utilizing deep autoencoders (DAEs) for automatic feature extraction and Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction.