DeepDFA: Automata Learning through Neural Probabilistic Relaxations
This work addresses the challenge of interpretable and efficient automata learning for applications in formal verification and grammar induction, representing a hybrid incremental advance.
The authors tackled the problem of learning Deterministic Finite Automata (DFA) from traces by introducing DeepDFA, a differentiable and discrete model that combines probabilistic relaxations with neural networks, resulting in improved accuracy, scalability, and noise resilience compared to traditional methods.
In this work, we introduce DeepDFA, a novel approach to identifying Deterministic Finite Automata (DFAs) from traces, harnessing a differentiable yet discrete model. Inspired by both the probabilistic relaxation of DFAs and Recurrent Neural Networks (RNNs), our model offers interpretability post-training, alongside reduced complexity and enhanced training efficiency compared to traditional RNNs. Moreover, by leveraging gradient-based optimization, our method surpasses combinatorial approaches in both scalability and noise resilience. Validation experiments conducted on target regular languages of varying size and complexity demonstrate that our approach is accurate, fast, and robust to noise in both the input symbols and the output labels of training data, integrating the strengths of both logical grammar induction and deep learning.