Test Case Prioritization with Z-Score Based Neuron Coverage

Deep neural networks (DNNs) have been widely used in various applications, such as autonomous driving, healthcare, etc. However, despite achieving high accuracy, DNNs have exhibited quality issues in various aspects such as vulnerability to data corruptions, adversarial attacks, and data dependencies. To ensure the integrity and reliability of these systems, the use of systematic verification and validation methodologies before the deployment of DNN is considered an indispensable technique. Test case prioritization techniques reduce the cost of DNN verification by prioritizing test cases that could induce mispredictions of DNN. In this paper, we propose a neuron coverage-based test case prioritization technique for the DNN classifier that assigns sample priorities based on the ratio of outlier-valued neurons among total neurons in DNN. We evaluate the proposed method with three publicly accessible datasets with different sizes of DNN. The experimental results demonstrate that the proposed method outperforms the existing state-of-the-art neuron coverage-based approach both in error-inducing sample prioritization effectiveness and inference time efficiency.