A Novel ResNet-Based Autoencoder Framework for Semi-Supervised Clustering Using Pairwise Constraints

Kadhim Khudhair, Noor; Balafar, Mohammad Ali; Tanha, Jafar; Golzari Oskouei, Amin

doi:10.22034/jasp.2025.64845.1263

A Novel ResNet-Based Autoencoder Framework for Semi-Supervised Clustering Using Pairwise Constraints

Document Type : Original Article

Authors

¹ Tabriz University

² Faculty of Electrical and Computer Engineering, University of Tabriz

³ Assistant Professor -Faculty of Electrical and Computer Engineering, University of Tabriz

⁴ Faculty of IT and Computer Engineering, Urmia University of Technology, Urmia, Iran

10.22034/jasp.2025.64845.1263

Abstract

Recently, the use of deep learning for data clustering has gained significant attention due to its ability to uncover complex structures within data. In this paper, a novel ResNet-based autoencoder framework for semi-supervised clustering is proposed. This framework utilizes an autoencoder architecture with residual connections, consisting of two main components: a ResNet-based encoder that extracts meaningful latent representations from the data, and a decoder that is responsible for accurately reconstructing the input data. The proposed model employs a composite loss function that integrates Mean Squared Error (MSE), Kullback-Leibler Divergence (KLD), semi-supervised pairwise constraints, and label-based loss. This innovative combination guides the clustering process using a target distribution as soft labels, ensuring the stability of the model. Experimental results on benchmark datasets demonstrate that the proposed model achieves an average clustering accuracy of 96.8% and 92.5% in Normalized Mutual Information (NMI). These results indicate a significant improvement in performance compared to existing methods in semi-supervised clustering.

Keywords

Main Subjects