ReSyn: A Generalized Recursive Regular Expression Synthesis Framework
a day ago
- #Divide-and-Conquer
- #Programming-by-Example
- #Regex Synthesis
- Existing PBE systems struggle with complex real-world regex patterns due to high structural complexity like deep nesting and unions.
- ReSyn is a synthesizer-agnostic divide-and-conquer framework that breaks down complex regex synthesis problems into manageable sub-problems.
- Set2Regex is introduced as a parameter-efficient synthesizer capturing permutation invariance of examples.
- Experimental results show ReSyn significantly improves accuracy across various synthesizers.
- Combining ReSyn with Set2Regex achieves new state-of-the-art performance on challenging real-world benchmarks.