The article presents Symbolic Machine Learning Prover (SMLP), a tool and library for system exploration based on data samples. SMLP uses a grey-box approach, combining statistical methods with machine learning models for system exploration. It has been used in an industrial setting at Intel for analyzing and optimizing hardware designs. The tool offers multiple capabilities for system design space exploration, including methods for selecting parameters for modeling design for configuration optimization and verification. The authors discuss the concept of ‘stability’ of an assignment to system parameters that satisfies all model constraints. The tool also deals with parameterized systems where parameters can be tuned to optimize system performance.
Publication date: 2 Feb 2024
Project Page: https://arxiv.org/abs/2402.01415
Paper: https://arxiv.org/pdf/2402.01415
