Auto-attention mechanism for multi-view deep embedding clustering

In several fields, deep learning has achieved tremendous success. Multi-view learning is a workable method for handling data from several sources. For clustering multi-view data, deep learning and multi-view learning are excellent options. However, a persistent challenge is a need for the current deep learning approach to independently drive divergent neural networks for different perspectives while working with multi-view data. The current methods use the number of viewpoints to calculate neural network statistics. Consequently, as the number of views rises, it results in a considerable calculation. Furthermore, they vainly try to unite various viewpoints at the training. Incorporating a triple fusion technique, this research suggests an innovative multi-view deep embedding clustering (MDEC) model. The suggested model can jointly acquire the specific knowledge in each view as well as the information fragment of the collective views. The main goal of the MDEC is to lower the errors made when learning the features of each view and correlating data from many views. To address the optimization problem, the MDEC model advises a suitable iterative updating approach. In testing modern deep learning and non-deep learning algorithms, the experimental study on small and large-scale multi-view data shows encouraging results for the MDEC model. In multi-view clustering, this work demonstrates the benefit of the deep learning-based approach over the non-ones. However, future work will address a variety of issues related to MDEC including the speed.