A Deep Embedded Clustering with Selective Data Augmentation
Mingcheng Zuo, Fujian Xu, Dunwei Gong · 2024
By facilitating unsupervised clustering with data augmentation, the clustering performance of deep embedded clustering has been significantly enhanced in DEC-DA. However, how to selectively augment valuable data and better present the distribution characteristics of the data is still a challenge. To fill the gap, this paper develops a deep embedded clustering with selective data augmentation. Firstly, the proposed method defines two typical sample distribution patterns. Then, the distribution pattern is determined by extracting the samples' distribution characteristics. Finally, according to the determined pattern, the samples are probabilistically selected and augmented under a possibility distribution function. The effectiveness and efficiency of the proposed algorithm are demonstrated by its competitive performance on UCR benchmarks and real-world applications.