A Survey on Test Input Selection and Prioritization for Deep Neural Networks

With the breakthrough advancements of deep neural network technology in applications such as image processing, autonomous driving, and speech recognition, the testing of deep neural network models becomes crucial to ensure their performance and reliability. The selection of test inputs is a critical step in the testing process. Prioritizing test inputs and selecting the most impactful ones can help improve testing efficiency to identify potential problems and deficiencies in the model as early as possible, given the large size of the test dataset and the high cost of annotation. In order to gain a deeper understanding of the research progress in the field of test input selection for deep neural networks, this paper conducts a survey of academic papers in the field over the past years. A systematic review of existing research outcomes is presented, focusing on the criteria, methods, and priority ranking for the selection of test inputs in deep neural networks. Additionally, the paper offers insights into future challenges for test input selection in deep neural networks.