Semi-supervised RVFL-based neural networks for solving classification problems

• Main Author: Yao, Cheng Hui
• Other Authors: Ponnuthurai Nagaratnam Suganthan
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/154134
• Institution: Nanyang Technological University

Random Vector Functional Link (RVFL) is widely used on supervised tasks. However, in the real world, we often have a small number of labelled samples and many unlabelled samples. In this paper, we extend RVFLs for semi-supervised tasks based on Manifold Regularization (MR), thus expanding on the application of RVFL to semi-supervised tasks. MR has been deeply researched in the past decade to improve the quality of classifiers making use of unlabelled data. Following this MR approach, semi-supervised RVFL (SS-RVFL) demonstrates great improvements in performance in comparison to typical RVFL networks. The method enhances RVFL based classifiers for semi-supervised learning while still retaining the efficiency of RVFL networks. In these experiments, we are also proposing the use of Deep RVFL networks for semisupervised learning. There have not been as much research regarding the use of semisupervised deep RVFL networks. Hence we will be applying the MR approach to Deep RVFL (dRVFL) and Ensemble Deep RVFL (edRVFL) for semi-supervised classification problems as well. Deep variants of the RVFL network are able to gain information from various enhanced patterns which could help in improving the performance of a learning algorithm. We present an evaluation on well-known datasets to demonstrate the performance of the proposed methods.