Learning based Widening

Even though design of a widening operator is an integral step in the design of an abstract interpreter using an infinite domain, not much work has been done to systematize the design of widening operator. While there exist works that derive widening of higher-level domains by lifting the widening of the base-level domain, the design of widening for baselevel domains remains largely creative process, especially for numerical domains. This is partly because the definition of widening requires convergence and termination while providing for very weak algebraic properties, making the problem of creating generic widening construction techniques hard. We provide a template widening operator, parametric on the shape of the domain, by leveraging a duality between Supervised Machine Learning (Regression) and Abstract Interpretation (Widening). CCS Concepts •Theory of computation→ Program analysis;