In the context of Continual Semantic Segmentation (CSS), replay-based methods tend to achieve better performance than knowledge distillation-based ones, as the former utilizes additional data to transfer old knowledge. However, this advantage is at the cost of necessitating additional space for storing the generative model and extra time for continual training. To address this predicament, we propose a novel CSS framework, namely Adversarial Attack-based Knowledge Retention (AAKR). The AKKR framework generates specific adversarial samples by adding images, and uses them to retain old knowledge. Specifically, we leverage adversarial attacks to generate adversarial images for incremental samples. By imposing additional constraints within these attacks, we enhance the transfer of old knowledge, thereby reinforcing the understanding of previously learned information. Furthermore, we design an attack probability module that adjusts adversarial attack directions based on training feedback. This module effectively encourages the new model to learn old knowledge from poorly protected classes, significantly improving knowledge transfer effectiveness. Our comprehensive experiments demonstrate the efficacy of AAKR, and showcase that AAKR surpasses state-of-the-art competitors on benchmark datasets.

Multimodal recommendation systems integrate diverse multimodal information into the feature representations of both items and users, thereby enabling a more comprehensive modeling of user preferences. However, existing methods are hindered by data sparsity and the inherent noise within multimodal data, which impedes the accurate capture of users' interest preferences. Additionally, discrepancies in the semantic representations of items across different modalities can adversely impact the prediction accuracy of recommendation models. To address these challenges, we introduce a novel diffusion-based contrastive learning (DiffCL) framework for multimodal recommendation. DiffCL employs a diffusion model (DM) to generate contrastive views that effectively mitigate the impact of noise during the contrastive learning phase. Furthermore, it improves semantic consistency across modalities by aligning distinct visual and textual semantic information through stable ID embeddings. Finally, the introduction of the item-item graph (I-I graph) enhances multimodal feature representations, thereby alleviating the adverse effects of data sparsity on the overall system performance. We conduct extensive experiments on three public datasets, and the results demonstrate the superiority and effectiveness of the DiffCL.

WebView is a UI widget that helps integrate web applications into the native context of Android apps. It provides powerful mechanisms for bi-directional interactions between the native-end (Java) and the web-end (JavaScript) of an Android app. However, these interaction mechanisms are complicated and have induced various types of bugs. To mitigate the problem, various techniques have been proposed to detect WebView-induced bugs via dynamic analysis, which heavily relies on executing tests to explore WebView behaviors. Unfortunately, these techniques either require manual effort or adopt random test generation approaches, which are not able to effectively explore diverse WebView behaviors. In this paper, we study the problem of test generation for WebViews in Android apps. Effective test generation for WebViews requires identifying the essential program properties to be covered by the generated tests. To this end, we propose WebView-specific properties to characterize WebView behaviors, and devise a cross-language dynamic analysis method to identify these properties. We develop ωTest, a test generation technique that searches for event sequences covering the identified WebView-specific properties. An evaluation on 74 real-world open-/closed-source Android apps shows that ωTest can cover diverse WebView behaviors and detect WebView-induced bugs effectively. ωTest detected 36 previously-unknown bugs. From the 22 bugs that we have reported to the app developers, 13 bugs were confirmed, 9 of which were fixed.

WebView is a widely used Android component that augments a native app with web browser capabilities. It eases the interactions between an app’s native code and web code. However, the interaction mechanism of WebView induces new types of bugs in Android apps. Understanding the characteristics and manifestation of these WebView-induced bugs (ωBugs for short) facilitates the correct usages of WebViews in Android apps. This motivates us to conduct the first empirical study on ωBugs based on those found in popular open-source Android apps. Our study identified the major root causes and consequences of ωBugs and made interesting observations that can be leveraged for detecting and diagnosing ωBugs. Based on the empirical study, we further propose an automated testing technique ωDroid to effectively expose ωBugs in Android apps. In our experiments, ωDroid successfully discovered 30 unique and previously-unknown ωBugs when applied to 146 open-source Android apps. We reported the 30 ωBugs to the corresponding app developers. Out of these 30 ωBugs, 14 were confirmed and 7 of them were fixed. This shows that ωDroid can effectively detect ωBugs that are of the developers’ concern.

Due to the availability of various open source Machine Learning (ML) tools and libraries, developers nowadays can easily implement their purposes by just invoking machine learning APIs without knowing the details of the algorithm. However, the owners of ML tools and libraries usually pay more attention to the correctness and functionality of their algorithm, while spending much less effort on maintaining their code and keeping their code at a high quality level. Considering the popularity of machine learning in today's world, low quality ML tools and libraries can have a huge impact on the software products that use ML algorithms. So in this paper, we conduct an empirical study on real machine learning bugs to examine their patterns and how they evolve over time. We collect three popular machine learning projects on Github, and manually analyzed 329 closed bugs from the perspectives of their bug category, fix pattern, fix scale, fix duration, and type of software maintenance. The results show that (1) there are seven categories of bugs in machine learning programs; (2) twelve different fix patterns are commonly used to fix the bugs; (3) 63.83% of the patches belong to micro-scale-fix and small-scale-fix, and 68.39% of the bugs are fixed within one month; (4) 47.77% of the bug fixes belong to corrective activity from the view of software maintenance.

In the open source community such as GitHub, developers usually need to find projects similar to their work, with the aim to reuse their functions and explore ideas of features that could be possibly added into their project at hand. Traditional text search engine can help detect similar resources. However, it is difficult for developers to use in open source community because a few query words cannot describe the whole features of a project. In this paper, we present a practical software recommendation system, REPERSP, which is used to recommend personalized software projects in GitHub. According to the features of projects created by developers and their behavior to other known projects, REPERSP recommends the top N relevant and personalized software projects. Moreover, REPERSP is implemented with the MapReduce parallel processing frame - Apache Spark for large-scale data, which can be scaled to a large number of users and projects for practical usage. Empirical results show that REPERSP can recommend more accurate results compared with other two recommendation algorithms, i.e., UserCF (user collaborative filtering) and ItemCF (item collaborative filtering). Video of our demo is available at https://youtu.be/WKigSUV4UA0.

Accurate and complete commit messages summarizing software changes are important to support various software maintenance activities. In practice, these commit messages are often manually submitted by individual software developer to provide information about the changes involved in the incremental changes. Hence, the content and quality of these commit messages may be different. For example, some commit messages are too short and lack of essential information while others with too much detailed information can be time-consuming to read. What's more, most of the commit messages focus on what has been changed by developers in a commit, but why they changed and the motivation behind the code changes (which can assist developers in understanding code changes), are usually ignored. In this paper, we present an approach that can automatically generate the commit messages related to the code changes, including not only what have been changed but also why they were changed. Our approach uses method stereotypes and the type of changes to generate commit messages. We evaluate our approach by comparing the quality of generated messages with the original commit messages written by the original developers and those generated by a state-of-art technique, i.e., ChangeScribe. The results demonstrate that the messages generated by our approach are preferred in about 69% of the cases.

Topic models are shown to be effective to mine unstructured software engineering (SE) data. In this paper, we give a simple survey of exploring topic models to support various SE tasks between 2003 and 2015. The survey results show that there is an increasing concern in this area. Among the SE tasks, source code comprehension and software history comprehension are the mostly studied, followed by software defects prediction. However, there is still only a few studies on other SE tasks, such as feature location and regression testing.

Software repositories such as revision control systems and bug tracking systems are usually used to manage the changes of software projects. During software maintenance and evolution, software developers and stakeholders need to investigate these repositories to identify what tasks were worked on in a particular time interval and how much effort was devoted to them. A typical way of mining software repositories is to use topic analysis models, e.g., Latent Dirichlet Allocation (LDA), to identify and organize the underlying structure in software documents to understand the evolution of development topics. These previously LDA-based topic analysis models can capture either changes on the strength (popularity) of various development topics over time (i.e., strength evolution) or changes in the content (the words that form the topic) of existing topics over time (i.e., content evolution). Unfortunately, few techniques can capture both strength and content evolution simultaneously. However, both pieces of information are necessary for developers to fully understand how software evolves. In this paper, we propose a novel approach to analyze commit messages within a project's lifetime to capture both strength and content evolution simultaneously via Online Latent Dirichlet Allocation (On-Line LDA). Moreover, the proposed approach also provides an efficient way to detect emerging topics in real development iteration when a new feature request arrives at a particular time, thus helping project stakeholds progress their projects smoothly.

As the development of a software project progresses, its complexity grows accordingly, making it difficult to understand and maintain. During software maintenance and evolution, software developers and stakeholders constantly shift their focus between different tasks and topics. They need to investigate into software repositories (e.g., revision control systems) to know what tasks have recently been worked on and how much effort has been devoted to them. For example, if an important new feature request is received, an amount of work that developers perform on ought to be relevant to the addition of the incoming feature. If this does not happen, project managers might wonder what kind of work developers are currently working on. Several topic analysis tools based on Latent Dirichlet Allocation (LDA) have been proposed to analyze information stored in software repositories to model software evolution, thus helping software stakeholders to be aware of the focus of development efforts at various time during software evolution. Previous LDA-based topic analysis tools can capture either changes on the strengths of various development topics over time (i.e., strength evolution) or changes in the content of existing topics over time (i.e., content evolution). Unfortunately, none of the existing techniques can capture both strength and content evolution. In this paper, we use Dynamic Topic Models (DTM) to analyze commit messages within a project's lifetime to capture both strength and content evolution simultaneously. We evaluate our approach by conducting a case study on commit messages of two well-known open source software systems, jEdit and PostgreSQL. The results show that our approach could capture not only how the strengths of various development topics change over time, but also how the content of each topic (i.e., words that form the topic) changes over time. Compared with existing topic analysis approaches, our approach can provide a more complete and valuable view of software evolution to help developers better understand the evolution of their projects.

With recent progress in vehicle autonomous driving and vehicular communication technologies, vehicle systems are developing towards fully connected and fully autonomous systems. This paper studies lane assignment strategies for connected autonomous vehicles in a highway scenario and their impact on the overall traffic efficiency and safety. We formulate a model of connected autonomous vehicles, which includes three features: traffic data available online, ultra-short reaction time, and cooperative driving. Based on this model, we propose a novel lane change maneuver Politely Change Lane (PCL), which achieves the tradeoff between traffic safety and efficiency. Its effectiveness is validated and evaluated by extensive simulations. The performance shows that PCL improves both safety and efficiency of the overall traffic, especially with heavy traffic.

We consider distributed top-k processing problem in a mobile scenario. Specially, we focus on a real application of a bus network (N nodes), where buses are equipped with cameras for real-time security surveillance. Due to the limited number of screens (k, k<;<;N) at the traffic management center, how to select k bus nodes with most passengers to upload image data by DSRC needs to be solved. We present a novel distributed scheme BSS to handle this so-called "top-k node selection" issue in bus network, in which challenges of low time cost and high accuracy is not trivial. BSS utilizes various strategies to speed up top-k node selection facing poor network condition and application requirements. Performance evaluation is carried out on a real-trace driven simulator, which utilizes about 700 buses in Shanghai. The testing results show that BSS has an excellent performance in terms of time cost and average degree of accuracy, which shows the effectiveness of BSS scheme for real-time security surveillance.