Fuzziness Based Random Vector Functional-Link Network for Semi-supervised Learning

To improve the generalization performance of random vector functional link networks (RVFL), we propose a novel fuzziness based RVFL algorithm (F-RVFL) from the perspective of fuzzy theory for semi-supervised learning. Compared with the RVFL algorithm, the proposed F-RVFL algorithm shows better generalization performance on classification problems. In addition, we studied the relationship between the samples' output fuzziness and the classifier performance and obtained some useful conclusions, which gives a new direction for RVFL performance optimization.