Abstract:Quality assurance of deep neural networks (DNNs) is crucial for the deployment of DNN-based software, especially in mission- and safety-critical tasks. Inspired by structural white-box testing in traditional software, many test criteria have been proposed to test DNNs, i.e., to exhibit erroneous behaviors by activating new test units that have not been covered, such as new neurons, values, and decision paths. Many studies have been done to evaluate the effectiveness of DNN test coverage criteria. However, existing empirical studies mainly focused on measuring the effectiveness of DNN test criteria for improving the adversarial robustness of DNNs, while ignoring the correctness property when testing DNNs. To fill in this gap, we conduct a comprehensive study on 11 structural coverage criteria, 6 widely-used image datasets, and 9 popular DNNs. We investigate the effectiveness of DNN coverage criteria over natural inputs from 4 aspects: (1) the correlation between test coverage and test diversity; (2) the effects of criteria parameters and target DNNs; (3) the effectiveness to prioritize in-distribution natural inputs that lead to erroneous behaviors; (4) the capability to detect out-of-distribution natural samples. Our findings include: (1) For measuring the diversity, coverage criteria considering the relationship between different neurons are more effective than coverage criteria that treat each neuron independently. For instance, the neuron-path criteria (i.e., SNPC and ANPC) show high correlation with test diversity, which is promising to measure test diversity for DNNs. (2) The hyper-parameters have a big influence on the effectiveness of criteria, especially those relevant to the granularity of test criteria. Meanwhile, the computational complexity is one of the important issues to be considered when designing deep learning test coverage criteria, especially for large-scale models. (3) Test criteria related to data distribution (i.e., LSA and DSA, SNAC, and NBC) can be used to prioritize both in-distribution natural faults and out-of-distribution inputs. Furthermore, for OOD detection, the boundary metrics (i.e., SNAC and NBC) are also effective indicators with lower computational costs and higher detection efficiency compared with LSA and DSA. These findings motivate follow-up research on scalable test coverage criteria that improve the correctness of DNNs.

Covering Algorithm Based on Neighborhood Search and Its Applications

Cuckoo Search Algorithm with Membrane Communication Mechanism for Modeling Overhead Crane Systems Using RBF Neural Networks.

A Sphere Neighborhood Covering-Structuring Approach Based on Neural Network and Its Application

A Modified Constructive Neural Networks And Its Application For Large-Scale Data Mining

Efficient Neighborhood Covering Reduction With Submodular Function Optimization

Coverage Algorithm of K-nearest Neighbor Based on Communication Beacon in Wireless Mobile Sensor Network

A Fuzzy Classifier Based on the Constructive Covering Approach in Neural Networks

A constructive learning algorithm for text categorization

A Multirobot Cooperative Area Coverage Search Algorithm Based on Bioinspired Neural Network in Unknown Environments

A Multi-Robot Cooperative Searching Algorithm in Unknown Environments Based on Neural Network

Adaptive Nearest Neighbor Algorithm with Dynamic Neighborhood

Learning Discriminative Features by Covering Local Geometric Space for Point Cloud Analysis.

A novel chaotic and neighborhood search-based artificial bee colony algorithm for solving optimization problems

The Localized Area Coverage Algorithm Based on Game-Theory for WSN.

Can Coverage Criteria Guide Failure Discovery for Image Classifiers? an Empirical Study

A novel neural network classification model based on covering and Affinity Propagation clustering algorithm

Subspace Learning of Neural Networks

Covering rough sets based on neighborhoods: An approach without using neighborhoods

Machine Learning Approaches for Developing Land Cover Mapping

A Novel Local Search Algorithm with Configuration Checking and Scoring Mechanism for the Set K-Covering Problem

A Cover Time Study of a non-Markovian Algorithm