Using Recurrent Neural Network to Predict Tactics for Proving Component Connector Properties in Coq

Formal modeling and verification of component connectors in complex software systems are getting more interests with recent advancements and evolution in modern software techniques. Various properties of connectors can be specified as high-order logic propositions and verified using theorem proving techniques. However, most high-order logic provers still highly rely on human interactions and thus make the proving process difficult and time-consuming. In this paper, we propose an approach based on recurrent neural networks (RNNs) to predict the correct tactics in the proving process. Recurrent layers consisting of Long-Short-Term-Memory (LSTM) units provide a better correctness rate comparing with simple RNN units. Under this framework, properties of connectors can be naturally formalized and semi-automatically proved in Coq.