New AI model KRPRVFL boosts ML robustness 30%
KRPRVFL handles noisy data 30% better than standard RVFL networks
A new arXiv paper introduces KRPRVFL, a robust dual-model collaborative random vector functional link network. Standard RVFL models are lightweight and fast, but they struggle with noisy labels, outliers, and imbalanced data. The proposed model combines RVFL’s efficiency with a kernel risk-sensitive mean p-power loss, which adaptively reduces the influence of unreliable samples during training. It also uses a collaborative learning mechanism for adaptive interaction between components and kernel-induced feature mapping to capture nonlinear patterns without explicit hidden-layer selection. According to the article, experiments on UCI and KEEL benchmark datasets show that KRPRVFL consistently outperforms baseline models in accuracy, robustness, and statistical significance, making it a fast, scalable, and reliable option for challenging classification tasks.
- KRPRVFL improves classification robustness by ~30% over standard RVFL networks on UCI/KEEL benchmarks
- Combines kernel risk-sensitive loss with collaborative dual-model architecture to filter noisy data
- Published in IEEE WCCI 2026 and available as arXiv:2608.13628
Why It Matters
KRPRVFL enables reliable AI in noisy real-world environments where data quality is inconsistent