Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore a 15-minute conference presentation that introduces a novel approach to Bayesian program analysis through abstraction refinement techniques. Learn how researchers from Peking University address the critical challenge of selecting program abstractions to effectively generalize from external information in Bayesian program analysis systems. Discover how conventional logical deduction in program analysis can be converted into Bayesian inference for improved accuracy by incorporating external knowledge. Examine the limitations of existing learning-based approaches that train selection policies on program datasets, which often result in suboptimal performance on new programs due to inadequate training set selection. Understand the innovative solution inspired by counterexample-guided refinement frameworks that searches for optimal abstractions dynamically during analysis. Master the application of conditional independence theory to refine abstractions and eliminate incorrect generalizations that compromise analysis accuracy. Analyze practical implementations through two concrete case studies: Bayesian thread-escape analysis and Bayesian datarace analysis, demonstrating significant performance improvements over existing methods. Access comprehensive supplementary materials including research artifacts and detailed experimental data to support further investigation into this cutting-edge intersection of machine learning and static program analysis.
