No abstract

With the increasing adoption of Deep Neural Network (DNN) models as integral parts of software systems, efficient operational testing of DNNs is much in demand to ensure these models' actual performance in field conditions. A challenge is that the testing often needs to produce precise results with a very limited budget for labeling data collected in field.Viewing software testing as a practice of reliability estimation through statistical sampling, we re-interpret the idea behind conventional structural coverages as conditioning for variance reduction. With this insight we propose an efficient DNN testing method based on the conditioning on the representation learned by the DNN model under testing. The representation is defined by the probability distribution of the output of neurons in the last hidden layer of the model. To sample from this high dimensional distribution in which the operational data are sparsely distributed, we design an algorithm leveraging cross entropy minimization.Experiments with various DNN models and datasets were conducted to evaluate the general efficiency of the approach. The results show that, compared with simple random sampling, this approach requires only about a half of labeled inputs to achieve the same level of precision. CCS CONCEPTS• Software and its engineering → Software testing and debugging; • Computing methodologies → Neural networks.

No abstract

No abstract

An additive genetic model is usually employed in case-control-based genome-wide association studies. The model usually encodes "AA", "Aa" and "aa" ("a" represents the minor allele) as three different numbers, implying the contribution of genotype "Aa" to the phenotype is different from "AA" and "aa". From the perspective of biological phenomena, the coding is reasonable since the phenotypes of lives are not "black and white". A case-control based study, however, has only two phenotypes, case and control, which means that the phenotypes are "black and white". It suggests that a recessive/dominant model may be an alternative to the additive model. In order to investigate whether the alternative is feasible, we conducted comparative experiments on several models used in those studies through chi-square test and logistic regression. Our simulation experiments demonstrate that a recessive model is better than the additive model. The area under the curve of the former has increased by 5% compared with the latter, the discrimination of identifying risk single nucleotide polymorphisms has been improved by 61%, and the precision has also reached 1.10 times that of the latter. Furthermore, the real data experiments show that the precision and area under the curve of the former are 16% and 20% higher than the latter respectively, and the area under the curve of dominant model of the former is 13% higher than the latter. The results indicate a recessive/dominant model may be an alternative to the additive model and suggest a new route for case-control-based studies.