EPIK: System-Level Observations for Bayesian Learning in Quantitative Verification
A novel technique known as EPIK has been developed to enhance the precision and reliability of quantitative software system verification by utilizing system-level observations to guide Bayesian parameter learning. This method tackles the issue of imprecise or unhelpful prior knowledge (PK) in Bayesian inference, which can result in erroneous verification outcomes regarding aspects such as reliability and response time. EPIK incorporates and integrates PK within quantitative verification through Bayesian estimators, differing from traditional methods that depend on PK related to formal model transition parameters; instead, it focuses on directly observable system-level properties linked to real-world semantics. The research, detailed in an arXiv paper (ID: 2608.03489v1), is pertinent to formal verification and probabilistic model checking.
Key facts
- EPIK is a new approach for Bayesian learning in quantitative verification.
- It leverages system-level observations to inform model parameters.
- It addresses the challenge of inaccurate prior knowledge in Bayesian inference.
- EPIK uses system-level properties that are directly observable.
- It formulates a twofold optimisation problem.
- The approach is described in arXiv paper 2608.03489v1.
- It aims to improve accuracy and robustness of verification results.
- It is relevant to software systems' reliability and response time analysis.
Entities
Institutions
- arXiv