Formal XAI via Syntax-Guided Synthesis

Katrine Bjørner, Samuel Judson, Filip Cano, Drew Goldman, Nicholas Shoemaker, Ruzica Piskac, Bettina Könighofer

Abstract: In this paper, we propose a novel application of syntax-guided synthesis to find symbolic representations of a model’s decision-making process, designed for easy comprehension and validation by humans. Our approach takes input-output samples from complex machine learning models, such as deep neural networks, and automatically derives interpretable mimic programs. A mimic program precisely imitates the behavior of an opaque model over the provided data. We discuss various types of grammars that are well-suited for computing mimic programs for tabular and image input data.