Verification of RNN-Based Neural Agent-Environment Systems


We introduce agent-environment systems where the agent is stateful and executing a ReLU recurrent neural network. We define and study their verification problem by providing equivalences of recurrent and feed-forward neural networks on bounded execution traces. We give a sound and complete procedure for their verification against properties specified in a simplified version of LTL on bounded executions. We present an implementation and discuss the experimental results obtained.

Proceedings of the 33rd AAAI Conference on Artificial Intelligence (AAAI 2019)