Optimization Deep Neural Network Test Case through Abductive Learning

The rapid advancement of deep neural network (DNN) models has enabled their widespread application across various domains, including face recognition and natural language processing. However, data-driven DNN models are prone to erroneous behavior when inadequately trained, necessitating extensive predictive labeling of test data to identify and mitigate defects. Manual labeling, however, remains both labor-intensive and inefficient. To address this limitation, several automated test predictions labeling methods have been proposed. These approaches, however, often result in a high rate of mislabeling, thereby limiting improvements in classification performance during model retraining.

To address these challenges, this study proposes a Deep Neural Network Test Case Optimization Method based on Abductive Learning. This method integrates domain knowledge with logical reasoning to optimize the labeling of incorrectly predicted samples and utilizes abductively labeled data to retrain DNN models, thereby reducing both human effort and time requirements. Experiments were conducted on four deep learning test sets using four classical neural network models and compared against eight baseline methods. The proposed method achieved a fault detection rate of 96.77%, surpassing baseline methods by up to 13.46%. Furthermore, the labeling accuracy within the selected test case subset reached a maximum of 98.99%.