A Novel RVFL-Based Algorithm Selection Approach for Software Model Checking

Weipeng Cao, Yuhao Wu, Qiang Wang, Jiyong Zhang, Xingjian Zhang, Meikang Qiu

Abstract: Software model checking is the technique that automatically verifies whether software meets the given correctness properties. In the past decades, a large number of model checking techniques and tools have been developed, reaching a point where modern model checkers are sophisticated enough to handle large-scale software systems. However, due to the fact that the software model checkering techniques are diverse and each of them is designed and optimized for a specific type of software system, it remains a hard problem for engineers to efficiently combine them to verify the complex software systems in practice. To alleviate this problem, we propose a novel algorithm selection approach based on Random Vector Functional Link net (RVFL) for software model checking, namely Kaleidoscopic RVFL (K-RVFL). The novel design of feature hybridization and fusion enables K-RVFL to extract more diverse and multi-level features. We have also carried out a thorough experimental evaluation on a publicly available data set and compared K-RVFL with a number of neural networks, including RVFL, Extreme Learning Machine (ELM), Stochastic Configuration Network (SCN), Back Propagation algorithm (BP), and Supporting Vector Machine (SVM). The experimental results demonstrate the usefulness and effectiveness of K-RVFL.