A Discriminant Correlation Neural Network for Feature Representation Learning

Abstract

The progress of multimedia computing, particularly through deep neural networks (DNNs), has gained remarkable attention across a wide array of research domains and practical applications, such as image classification, video computing, natural language processing, among others. However, the inherent black-box characteristics and concern for sustainability of DNNs have emerged as significant challenges, especially when dealing with multi-view data. To tackle these challenges, alternative learning models have been actively explored in recent years, with green learning and interpretable learning standing out as two key directions. In this study, we explore one such model aimed at two-view feature representation learning. This proposed model leverages a perceptron-style neural network grounded in discriminant correlation analysis, which is named DC-PNN. The model incorporates statistical machine learning (SML) techniques to optimize network training. To evaluate its effectiveness and generalizability, we conduct experiments on audio emotion recognition and action recognition on two public data sets (i.e., RML emotional database and NTU RGB+D 120 database). Then, move forward to a real application, Glaucoma image classification. Experimental results demonstrate that the DC-PNN model outperforms state-of-the-art (SOTA) methods. In addition, the parameters and FLOPs analysis are implemented, validating the advantage of the model in terms of efficiency and resource requirements.