Adversarial training is a widely used technique to enhance the robustness of deep learning models. However, it is computationally expensive due to the need to calculate gradients for generating adversarial samples. While fast gradient sign method (FGSM)-based adversarial training is commonly employed to improve model robustness, it still incurs significant training time and computational overhead compared to standard training. Moreover, the robustness of models trained with FGSM-based adversarial training often remains unsatisfactory. Among the most efficient training techniques, mixed precision-based training reduces computational overheads by accelerating the standard training process. However, applying mixed precision to adversarial training does not reduce computational overheads and increase robustness. Thus, this article proposes a novel adaptive precision layering-based adversarial training and gradient accumulation method to reduce the computational cost. The proposed method efficiently assigns precision levels ( single and half precision ) to each model layer based on its sensitivity to precision loss. The less sensitive layers are assigned with half-precision, and the output layer is assigned with single precision. The perturbations are generated during the adversarial training by leveraging the fast gradient sign method with a random initialization technique. Additionally, the accumulation steps are also adaptively set based on the gradient flow and stability analysis. The adaptive precision layering accelerates the utilization of the graphics processing unit, and the efficient gradient accumulation method avoids unnecessary gradient calculation, thus reducing training time. By adaptively assigning the precision, gradient accumulation, and random initialization, we effectively solve three major problems in adversarial training: 1) significantly reduce the training time overheads, 2) solve the catastrophic overfitting, and 3) improve the robustness of the target models. Experimental evaluations on three state-of-the-art self-driving car models demonstrate an excellent improvement in adversarial robustness while significantly reducing the computational overheads.

Deep learning models have proven their effectiveness in intelligent transportation. However, their vulnerability to adversarial attacks poses significant challenges to traffic safety. Therefore, this article presents a novel technique to evaluate and improve the adversarial robustness of the deep learning models. We first proposed a deep-convolutional-autoencoder-based adversarial attack detector that identifies whether or not the input samples are adversarial. It serves as a preliminary step toward adversarial attack mitigation. Second, we developed a conditional generative adversarial neural network (c-GAN) to transform the adversarial images back to their original form to alleviate the adversarial attacks by restoring the integrity of perturbed images. We present a case study on the traffic sign recognition model to validate our approach. The experimental results showed the effectiveness of the adversarial attack mitigator, achieving an average structure similarity index measure of 0.43 on the learning interpretable skills abstractions (LISA)-convolutional neural network (CNN) dataset and 0.38 on the German traffic sign recognition benchmark (GTSRB)-CNN dataset. While evaluating the peak signal noise ratio, the c-GAN model attains an average of 18.65 on the LISA-CNN and 18.05 on the GTSRB-CNN dataset. Ultimately, the proposed method significantly enhanced the average detection accuracy of adversarial traffic signs, elevating it from 72.66% to 98% on the LISA-CNN dataset. In addition, an average of 28% improvement in accuracy was observed on the GTSRB-CNN.

Abstract A large volume of highly diverse and real‐time data provided by social network sites has revolutionized several data analytical sciences such as stock exchange fluctuation prediction, sentiment analysis, and political affiliations prediction. From a software engineering viewpoint, social network sites have produced an unprecedented opportunity for software development organizations to monitor the opinion of the large population of users of their systems. The extracted data from social network sites can be used to extract information about newly launched systems, and it enable software development organizations to obtain rapid technical and social feedback about their systems. This paper presents on a systematic literature review of the published research articles on how requirements engineering (RE) activities are conducted using social network sites, elaborated in studies published between 2011 and 2019. From the resulting studies, we extracted data to answer our four questions related to social network site (SNS)‐based RE approaches focusing on requirements elicitation, prioritization, and negotiation activities. Thirty‐one relevant studies were identified and analyzed to answer our research questions. Among 31 studies, 62% of the studies reportedly support requirements elicitation activity, and 14% of the studies reportedly support requirements elicitation, prioritization, and negotiation activities. Ten percent studies were about requirements elicitation and prioritization activities, 7% of the studies support both requirements elicitation and negotiation activities, and prioritization activity was supported by 7% of the studies. It was also observed that 58% of the selected studies adopted semi‐automatic approaches to conduct RE activities while 21% of the selected studies proposed automatic tools, and 21% of the selected studies adopted manual approaches to conduct RE activities. The outcomes of this systematic literature review indicate that social network sites can be a major source that can be used successfully to extract and identify user requirements. Moreover, this systematic literature review also presents implications of this study for practitioners and researchers including challenges, unresolved issues, and future research directions.

Abstract Large software development organizations manage reusable software components through a reusable software repository in order to reduce development time and cost and to improve software quality and productivity. This paper presents a hybrid DevOps process with a systematic reuse‐based software development and management process to reduce the effort and cost required for the rework and to increase productivity. The proposed approach promotes the systematic reuse of software components based on both information retrieval and ontology‐based retrieval techniques. The reusable assets are presented in different styles to ease and support the reuse process with fine‐grained reusable artifacts. To evaluate our proposed process, a pilot project, aiming to monitor the health of a patient, was developed and monitored the reuse activities throughout the whole experiment. The results revealed that our proposed process got an average gain of 35.2% in terms of developed function points by reusing 30.63% of reusable artifacts available in the reuse repository.

Due to diverse set of heterogeneous computing devices communicating with one another and fusing with physical components in Cyber-Physical Systems, software engineers may use different tools and/or modeling languages to formally describe or verify the system properties. As a result, the integration of these diverse constituents poses key challenges such as task for identifying interactions of components to be synthesized for a function in the systems. Although existing studies such as ontology and integration semantic languages have been used for specifying interactions of components in a Cyber-Physical System, these are still not applicable to discover the component interactions in collaborative Cyber-Physical Systems. It is due to the fact that functionalities of Cyber-Physical Systems are generally realized through interactions among multiple systems in a collaborative environment. This paper proposes a model interaction language, CyPhyML+ which can identify component interactions of realized functions in collaborative Cyber-Physical Systems. We show the proposed approach validity and applicability via an Automatic Incident Detection System.

When defects in safety-critical systems are exposed to actual mishaps, this can lead to property loss, injury, and/or death. Therefore, safety analysis should be performed to identify and minimize potential faults at the design phase of the software development life cycle. However, cyber-physical systems, in which multiple systems are interconnected by the network, may still pose potential risks because the composite analysis of cyber-physical systems is yet to be addressed; even though the hazard analysis for a single system is performed but it is insufficient to analyze hazards for cyber-physical systems. This paper presents an approach to support fault traceability by identifying content relationships among applied hazard analysis techniques to a cyber-physical system. The applied analysis techniques are Fault Tree Analysis, Event Tree Analysis, and Failure Mode and Effect Analysis. In this work, the relationships among these hazard analysis techniques are identified, and a Fault Traceability Graph is designed to represent the relationship among them. Our proposed approach enables to determine the propagation scope of a fault in order to find the source of a fault. Lastly, we propose a Fault Traceability Matrix to trace the faults and to make sure that all faults are controlled by the safety guards. As a result, we can effectively manage and control the faults at the design phase of the cyber-physical system to ensure the development of a safe system. We applied our proposed approach to Automatic Incident Detection System a real case study to show its validity and applicability.

Requirements traceability helps to ensure that a re- quirements specification is aligned with the intended stakehold- ers' needs. Such alignment should involve the consideration of why such needs arise, in terms of what problems the stakeholders are faced with, and what kinds of software system may help alle- viate or eliminate the problems. However, little work can be found on requirements traceability that explicitly considers the problems. In this paper, we propose an problem-aware frame- work for establishing requirements traceability, in the context of goal-oriented requirements engineering, which explicitly models problems and their root causes, together with other important ontological concepts, stakeholders' goals, and both functional and non-functional requirements as a solution, and issues. In this framework, ontological concepts are partitioned into layers, re- flecting which traceability links are classified into intra- and in- ter-traceabilities leading to several kinds of links. Additionally, undesirable consequences of inappropriate traceabilities are also categorized. A case study shows some key benefits of the frame- work.

Compressed video bitstreams are very sensitive to transmission errors. If we lose packets or receive them with errors during transmission, not only the current frame will be corrupted, but also the error will propagate to succeeding frames due to the spatiotemporal predictive coding structure. Error detection and concealment is a good approach to reduce the bad influence on the reconstructed visual quality. To increase concealment efficiency, we need to get some more accurate error detection algorithm. In this paper, we present a new error detection scheme based on a fragile watermarking algorithm to increase the error detection ratio. We verified that the pro posed algorithm generates good performances in PSNR and objective visual quality through the computer simulation by H.324M mobile simulation set.

During the development and maintenance of object-oriented (OO) software, the information on the classes which are more prone to be changed is very useful. Developers and maintainers can make a more flexible software by modifying the part of classes which are sensitive to changes. Traditionally, most change-proneness prediction has been studied based on source codes. However, change-proneness prediction in the early phase of software development can provide an easier way for developing a stable software by modifying the current design or choosing alternative designs before implementation. To address this need, we present a systematic method for calculating the behavioral dependency measure (BDM) which helps to predict change-proneness in UML 2.0 models. The proposed measure has been evaluated on a multi-version medium size open-source project namely JFreeChart. The obtained results show that the BDM is an useful indicator and can be complementary to existing OO metrics for change-proneness prediction.

Almost methods in embedded software modeling take software architecture as one of the crucial parts of the design process. Since a principal role of embedded software is the interaction with its environments, architectures that describe the dynamic aspects are also needed. Although such architectural view is significantly important in embedded software development, the existing approaches do not practically guide the architectural design based on the dynamic aspects. In this paper, we propose an operational architecture to represent the dynamic aspects of an embedded software. Also we explain how to use our operational architecture in the process of embedded software modeling. Our operational architecture effectively reflects interaction-based dynamic behavior of the system, and gives traceability between requirements and design model.

Many prior approaches in UML-based embedded software design incorporate state-based behavior modeling. However, interaction-based behavior modeling provides more intuitive view of a system. In this paper, we propose an approach to interaction-based behavior modeling of embedded software using UML 2.0. We use the interaction overview diagrams and the sequence diagrams to model the behavior. We present the method of constructing interaction-based behavior model with an example. We also briefly describe the idea of generating executable code from it.

Scenario-based modeling is emerging as a key technology in software development because scenarios are easy to describe and understand, naturally describing the real world, and highly used in several development phases. Scenario-Based Software Modeling (SBSM) represents scenarios with a formal method, and then integrates each scenario into a global one excluding redundant, incomplete, and inconsistent elements. An issue which is not yet proposed a clear solution in SBSM is the handling of different abstraction levels. In the existing proposals, this problem is hidden with an implicit assumption that all such scenarios have the same abstraction levels, which we believe to be unrealistic. In our research, we propose a Hierarchical Object-Oriented Petri Net (HOONet) as a method to specify the scenarios, and also suggest a method to integrate scenarios including different abstraction levels as well as redundancy, incompleteness, and inconsistency. Coordinating abstraction levels of scenarios can be achieved by considering scenario models as state-based transition models and verifying the transition models with respect to behavioral equivalence. In order to show the effectiveness of our suggestion, overall processes of SBSM using HOONet are explained with a practical example.

Petri net is used widely to analyze and model various systems formally. Recently, many Petri nets mania devote their efforts to enhancing and extending the expressive power of Petri nets. One such effort is to extend Petri nets with object-oriented concepts. An object-oriented paradigm provides excellent concepts to model real-world problems. Object-oriented concepts allow us to build software systems easily, intuitively, and naturally. Although several high-level Petri nets with the concept of objects are suggested, these nets do not fully support the object-oriented concepts. In this paper, we propose a hierarchical object-oriented Petri net (HOONet). The formal syntax and semantics of HOONet are explained in detail. HOONet supports a wide range of object-oriented features including abstract, encapsulated and modularized objects, object interaction by message passing, inheritance, and ploymorphism. HOONet also supports a variety of modeling and analysis mechanisms such as incremental modeling of evolving systems, unfolding the HOONet to lower level Petri net, and incremental reachability analysis for HOONet models. We demonstrate the usefulness of HOONet by applying it to modeling and analysis with an example.

Although object-oriented paradigm is becoming a more realistic approach to the development of large-scale software systems, the existing object-oriented notations and methodologies do not fully support the development of distributed object systems. In this paper, we integrate Meta-Object Protocol (MOP) into a de facto standard object-oriented modeling language UML together to build a software architecture for distributed object systems. We propose a high-level extension of conventional MOPs, called diMOP which helps to develop distributed object systems by realizing a reflective architecture. To incorporate diMOP with UML, we introduce two new specification languages: Class Diagram Supporting diMOP (CDSM) and Dynamically Configurable Object-oriented Statemachine (DCOS), which are proposed to replace the class diagram and the state diagram of UML. The two specification languages support the specification of dynamic configuration behaviors as well as incorporating the diMOP. This paper gives a methodology to develop efficiently distributed object systems through UML.