Learning Abstraction Selection for Bayesian Program Analysis
summary
Framework learns which abstraction parameters to change by extracting derivation graphs before and after a candidate refinement, computing graph features (count of reachable vertices, average shortest distance, count within radius K) and deciding refinements with an offline classifier; it turns the chosen abstraction into a Bayesian network with probabilities on derivation steps (example 0.8) and ranks alarms. Evaluated on data‑race and thread‑escape analyses over 13 Java programs, it reduces simulated interaction inversion count by about 40% and ships reusable artifacts.
AI-written summary · about
chapters (8)
- 0:00Bayesian program analysis intro
- 2:27Abstraction selection challenges
- 8:16Bingraph motivation
- 9:20Bingraph abstraction learning
- 11:29Bingraph feature calculation
- 13:14Experimental results
- 15:25Conclusion and future work
- 16:11Q&A
related talks
15mOn Abstraction Refinement for Bayesian Program Analysis
13mTowards Compiling Higher-Order Programs to Bayesian Networks
16mCombining Formal and Informal Information in Bayesian Program Analysis via Soft(…)
15mAbstraction Refinement-guided Program Synthesis for Robot Learning from Demonstrations
18mActive Learning for Neurosymbolic Program Synthesis
24mFuzzing Guided by Bayesian Program Analysis
18mTaming Reachability Analysis of DNN-Controlled Systems via Abstraction-Based Tr...
31mLinear Algebraic Abduction with Partial Evaluation