Deep Multi-scale Subspace Clustering Based on Bayesian Optimization

Deep subspace clustering leverages deep neural networks to learn low-dimensional feature representations of raw data, constructs a similarity matrix between data points, and employs spectral clustering to obtain the final clustering results. However, in the real world, data often faces challenges such as high dimensionality and complex structures, and obtaining more robust data representations to enhance the performance of deep subspace clustering remains a challenge. At the same time, Existing algorithms also suffer from manual hyperparameter tuning, affecting their universality and efficiency. To address these issues, this paper proposes a Deep Multi-Scale Subspace Clustering algorithm based on Bayesian Optimization (DMSSC-BO). It employs a multi-scale autoencoder for feature extraction and reconstruction, optimizing feature representation. The algorithm employs a pre-training and fine-tuning approach, with Bayesian optimization automating the search for optimal hyperparameters, thus reducing time and computational costs. Tests on standard and brain tumor datasets show that our approach improves clustering performance, yielding competitive results.