Decision Tree Learning in CEGIS-Based Termination Analysis

Abstract We present a novel decision tree-based synthesis algorithm of ranking functions for verifying program termination. Our is integrated into the workflow CounterExample Guided Inductive Synthesis (CEGIS). CEGIS an iterative learning model where, at each iteration, (1) synthesizer synthesizes candidate solution from current examples, and (2) validator accepts if it correct, or rejects providing counterexamples as part next examples. main novelty in design synthesizer: building on top usual tree algorithm, our detects cycles set example transitions uses them refining trees. have implemented proposed method obtained promising experimental results existing benchmark sets (non-)termination verification problems that require piecewise-defined lexicographic affine functions.