Software project defect prediction can help developers allocate debugging resources. Existing software defect prediction models are usually based on machine learning methods, especially deep learning. Deep learning‐based methods tend to build end‐to‐end models that directly use source code‐based abstract syntax trees (ASTs) as input. They do not pay enough attention to the front‐end data representation. In this paper, we propose a new framework to represent source code called multiperspective tree embedding (MPT‐embedding), which is an unsupervised representation learning method. MPT‐embedding parses the nodes of ASTs from multiple perspectives and encodes the structural information of a tree into a vector sequence. Experiments on both cross‐project defect prediction (CPDP) and within‐project defect prediction (WPDP) show that, on average, MPT‐embedding provides improvements over the state‐of‐the‐art method.

No abstract

No abstract

Software bug localization is very important in software engineering, but it is also complicated and time consuming. To improve the efficiency of developers, researchers have developed various traditional bug localization and machine learning bug localization methods. In this paper, we propose a novel method that improves bug localization performance. First, surface lexical correlation matching between bug reports and source code files is used to obtain features by deep neural network. Second, to solve the lexical gap between bug reports and source code files, semantic correlation matching between them is used to obtain features based on word embedding and sentence embedding by deep neural network. Then, the joint features obtained by the surface lexical and semantic correlation matching are fused into a unified feature representation for bug reports and source code files. In addition, since our experimental datasets are imbalanced data, we use a focal loss function to solve the impact of data imbalance. Finally, our method obtains the relatively high bug localization performance compared to other classic methods.

Although software bug localization in software maintenance and evolution is cumbersome and time-consuming, it is also very important, especially for large-scale software projects. To lighten the workload of developers, researchers have developed various information retrieval (IR)-based bug localization models for automated software support. In this paper, we propose a new method that reduces the time required for bug localization. First, the surface lexical similarity between a bug report and source code file is calculated based on the vector space model. Second, to address the lexical gap between the programming language and natural language, the word vector is used to calculate the semantic similarity between the bug report and source code file. Then, we use surface lexical and semantic similarity to calculate the total similarity for detecting buggy source code files. Our experimental word vectors are derived from Skip-gram and GloVe model training. We select an optimal 100 dimensional word vector for bug localization by evaluating it on four open source software examples. Finally, our experimental results show that our method outperforms classical IR-based methods in locating relevant source code files based on several indicators.

No abstract

As the convolutional neural network (CNN) gets deeper and wider in recent years, the requirements for the amount of data and hardware resources have gradually increased. Meanwhile, CNN also reveals salient redundancy in several tasks. The existing magnitude-based pruning methods are efficient, but the performance of the compressed network is unpredictable. While the accuracy loss after pruning based on the structure sensitivity is relatively slight, the process is time-consuming and the algorithm complexity is notable. In this article, we propose a novel automatic channel pruning method (ACP). Specifically, we firstly perform layer-wise channel clustering via the similarity of the feature maps to perform preliminary pruning on the network. Then a population initialization method is introduced to transform the pruned structure into a candidate population. Finally, we conduct searching and optimizing iteratively based on the particle swarm optimization (PSO) to find the optimal compressed structure. The compact network is then retrained to mitigate the accuracy loss from pruning. Our method is evaluated against several state-of-the-art CNNs on three different classification datasets CIFAR-10/100 and ILSVRC-2012. On the ILSVRC-2012, when removing 64.36% parameters and 63.34% floatingpoint operations (FLOPs) of ResNet-50, the Top-1 and Top-5 accuracy drop are less than 0.9%. Moreover, we demonstrate that without harming overall performance it is possible to compress SSD by more than 50% on the target detection dataset PASCAL VOC. It further verifies that the proposed method can also be applied to other CNNs and application scenarios.

No abstract