Research on Neural Network Fuzzy Testing Method Based on Reinforcement Learning Algorithm
Existing neural network fuzzy testing techniques usually adopt random mutations to initial samples during the test sam-ple generation phase,resulting in the low quality of generated samples and testing coverage.To address these issues,a neural net-work fuzzy testing technique based on reinforcement learning algorithm is proposed.Taking the fuzzy testing process as a Markov de-cision process,testing samples are regarded as environmental states in the model,different mutation methods as selected action space,and neuron coverage serves as a reward feedback,reinforcement learning algorithms are used to learn optimal mutation strategy,guid-ing the generation of optimal test samples to achieve the highest neuron coverage.Compared with the experiments for mainstream neural network fuzzy testing methods,the results show that the neural network fuzz testing technique based on reinforcement learning algorithm can improve the neuron coverage of different granularities.
fuzzy testneural networkreinforcement learningMarkov decision processreward function
万方数据
维普
