GB-RVFL: Fusion of Randomized Neural Network and Granular Ball Computing

TL;DR

This work proposes graph embedding GB-RVFL (GE-GB-RVFL) model, which fuses granular computing and graph embedding (GE) to preserve the topological structure of GBs and enhances scalability by requiring only the inverse of the GB center matrix.

Abstract

The random vector functional link (RVFL) network is a prominent classification model with strong generalization ability. However, RVFL treats all samples uniformly, ignoring whether they are pure or noisy, and its scalability is limited due to the need for inverting the entire training matrix. To address these issues, we propose granular ball RVFL (GB-RVFL) model, which uses granular balls (GBs) as inputs instead of training samples. This approach enhances scalability by requiring only the inverse of the GB center matrix and improves robustness against noise and outliers through the coarse granularity of GBs. Furthermore, RVFL overlooks the dataset's geometric structure. To address this, we propose graph embedding GB-RVFL (GE-GB-RVFL) model, which fuses granular computing and graph embedding (GE) to preserve the topological structure of GBs. The proposed GB-RVFL and GE-GB-RVFL models are evaluated on KEEL, UCI, NDC and biomedical datasets, demonstrating superior performance compared to baseline models.

Figures (12)

• Figure 1: Granular ball generation process.
• Figure 2: The visualization of generating granular balls by splitting the "fourclass" dataset. It assigns the label "$+1$" to the green granular balls, while the label "$-1$" is used for the magenta granular balls.
• Figure 3: Visual representation depicting the framework of the RVFL.
• Figure 4: Effect of different labels of noise on the performance of the proposed GB-RVFL and GE-GB-RVFL models.
• Figure 5: Effect of different labels of noise on the performance of the proposed GB-RVFL and GE-GB-RVFL model with the baseline IF-RVFL and NF-RVFL models.