VeriAbsL: Scalable Verification by Abstraction and Strategy Prediction (Competition Contribution)
Priyanka Darke, Bharti Chimdyalwar, Sakshi Agrawal, Shrawan Kumar, R. Venkatesh, Supratik Chakraborty
Abstract
Abstract We present VeriAbsL, a reachability verifier that performs verification in three stages. First, it slices the input code using a combination of two slicers, then it verifies the slices using predicted strategies, and at last, it composes the result of verifying the individual slices. We introduce a novel shallow slicing technique that uses variable reference information of the program, and data and control dependencies of the entry function to generate slices. We also introduce a novel strategy prediction technique that uses machine learning to predict a strategy. It uses boolean features to describe a program to a neural network that predicts a strategy. We use the portfolio of VeriAbs, a reachabiltiy verifier with manually defined strategies. In sv-comp 2023, VeriAbsL verified 227 (Without witness validation.) more programs than VeriAbs, and 475 (Without witness validation.) programs that VeriAbs could not verify.