As deep neural network (DNN) has made significant advancements across various fields, systematically testing DNN has become increasingly crucial. To uncover potential faults within DNN, a vast number of test cases and their correct labels are required, but the process of labeling is time-consuming and labor-intensive. To alleviate the burden on developers, test case selection techniques for DNN models have been proposed, assisting in the selection of test cases from large datasets that are more likely to reveal model faults. In this study, we introduce an efficient adaptive test case selection method based on the principle of uniform distribution of test cases, named EATS. In addition, we propose a test case optimization method and an image similarity calculation method. The optimization method can save time during the test case selection process, while the image similarity calculation method computes the degree of difference between images based on model uncertainty. EATS leverages the model’s uncertainty to achieve a more uniform distribution of selected test cases, aiming to select test cases that can induce a diversity of model fault predictions, thereby optimizing model performance. We conducted comparative experiments of EATS and other test case selection strategies on four common datasets and their corresponding DNN models. The experimental results show that EATS outperforms other methods in terms of the uniformity of test case distribution, diversity of errors discovered, and model optimization. It also demonstrates excellent time efficiency.

The widespread application of deep learning (DL) makes it crucial to ensure its reliability. Mutation testing has been employed in DL testing to evaluate the quality of test suite. However, the largest problem of DL mutation testing is the high cost of executing numbers of mutants. Weak mutation technology can alleviate this problem by reducing the execution time of mutants in traditional software testing. However, the compared components in traditional software are too trivial to apply weak mutation to DL models directly for that it is impractical for testers to track and monitor massive parameters during execution process. In this paper, we propose a novel weak mutation framework for mutants generated by source-level mutation operators. DeepWeak treats all layers that make up the DL model directly as a set of components of model to replace trivial parameters. And it pays attention to the last convolutioanl layer for that they not only have impacts on prediction results but also are evident for weak analysis. By quantifying contribution of feature maps to the prediction, weight maps will be obtained on the basis of their weights. Finally, the judgements on whether mutants have been killed will be reached by comparing the maps. To evaluate the applicability and effectiveness of our approach, we conduct experiments on three widely used datasets and four deep learning models using three metrics. Experimental results show that DeepWeak is effective at alleviating costs problem, reducing runtime by 11.21% to 18.21% compared with the DL mutation testing with little accuracy loss.

A novel study of ultra wideband (UWB) communication over the indoor powerline channel, in a wide frequency range up to 1 GHz is presented. An exhaustive measurement campaign was conducted on a test bed that replicated the environment of an indoor powerline network. The aim of this study was to observe and analyse the transmission and noise properties of such a broadband powerline channel. A time domain channel model has been used to study the broadband channel response of the powerline. Measurement and modelling results show that the indoor powerline channel provides a communication link in the 50–550 MHz frequency range. Channel capacity is greatly enhanced by exploiting the 500 MHz bandwidth. The conclusions of this study indicate that gigabit per second data rate transmission are possible over the indoor powerline channel in 50–550 MHz.