SCICO

Mansab et al. introduce TNA-CAF, a vulnerability detector that aligns code tokens with AST nodes via constrained cross-attention. It achieves strong F1/ROC-AUC on 195K+ C/C++ functions and shows strong cross-lingual generalization to Python. authors.elsevier.com/a/1n2nic7X5IcF8

Liu et al. introduce CodeRAG, a retrieval-augmented multimodal framework for software vulnerability detection, using a lightweight Transformer followed by an LLM to handle unclear cases. It improved F1/recall over baselines on Devign and Reveal datasets. authors.elsevier.com/a/1m-2Lc7X5IcEy

Alwadaeen and Alzamil introduce Optimized Greedy Additional (OGA), a faster test-case prioritization method using lazy heap updates, bitmaps, and similarity filtering. It achieved large speedups with negligible fault-detection loss on 54 Java projects authors.elsevier.com/a/1m%7EfXc7X5I…

Manfred Broy introduces a formalism for nondeterministic distributed systems beyond monotonicity, using timed streams to model feedback loops and concurrency. The paper also discusses the practical significance and applicability of the formalism. doi.org/10.1016/j.scic…

Arcaini and Cetinkaya introduce DETOUR, a regression testing tool for ADAs that uses hierarchical clustering to prioritize/select road-structure tests that cause lane departures. It is demonstrated with an illustrative example in its prioritization mode. authors.elsevier.com/a/1m%7ERVc7X5I…

Trovato et al. introduce AMBER, a JMH extension that classifies time series to detect steady-state execution and dynamically stop warm-up iterations. It was evaluated across over 30 Java projects, demonstrating higher accuracy and shorter test times. doi.org/10.1016/j.scic…

Leonhard and Stocco introduce PerturbationDrive for perturbation-based testing of ADAS, applying 30+ image perturbations. PerturbationDrive supports both offline evaluation on static datasets and online closed-loop testing in different simulators. doi.org/10.1016/j.scic…

Chouikh et al. introduce a rule-based co-evolution approach that uses ATL transformations from metamodel differences for semantic preservation in Maude. It is evaluated on a UML Sequence Diagram, showing reduced repetitive manual adaptations from users. authors.elsevier.com/a/1mv73c7X5IcCy

Mock et al. introduce Beyond Balance, a loss-weighting framework to address class imbalance in fine-tuning Transformer models for code comment classification. It consistently improves F1 scores across datasets, showing effective handling of rare classes. doi.org/10.1016/j.scic…

Chondamrongkul et al. introduce RepoAI, a model-agnostic multi-agent framework for automated code refactoring using LLM orchestration and RAG. It is evaluated on 1000+ RefactorBench tasks across multiple models, showing gains over direct generation. authors.elsevier.com/a/1mtMEc7X5IcCm

Pizard and Marenco introduce a scenario-based training approach to help learners apply systematic review evidence in software engineering decisions. This is evaluated via an exploratory mixed-methods study across three course iterations. authors.elsevier.com/a/1msAAc7X5IcCZ

Ma et al. introduce MutDBD, a mutation-based framework that identifies attacked classes in backdoored DNNs. It was evaluated across 32 scenarios on CIFAR-10, GTSRB, and ImageNet, showing reduced attack success rates while maintaining model accuracy. authors.elsevier.com/a/1mtApc7X5IcCa

Kertusha et al. present an SLR on web testing, highlighting trends, tools, and the growing role of AI-driven. The survey reveals limited industrial adoption and human-centered validation despite increasing use of real-world systems. doi.org/10.1016/j.scic…

Abe and Hayashi introduce ICCheck, a language-agnostic tool that synchronizes code clones. It was evaluated across multiple repositories, demonstrating fast (median 0.27s) detection of overlooked clone inconsistencies in 30+ languages. doi.org/10.1016/j.scic…

Broy introduces a theoretical framework for modeling concurrent distributed systems to handle non-deterministic and non-monotonic behavior. It shows how timed streams serve as an auxiliary construct for defining concurrent composition with feedback. sciencedirect.com/science/articl…

Xu et al. introduce change metrics that use runtime differences between pretraining and fine-tuning to identify bugs in fine-tuning. It evaluates it by training 5 learners across 6 datasets and 233 models, showing improved fault diagnosis over baselines. authors.elsevier.com/a/1mou7c7X5IcBv

Chen et al. introduce SLVD-CC, which combines vulnerability-aware value-flow slicing on LLVM IR (in SSA form) with an ordered message-passing GNN. It is evaluated on SARD and NVD, achieving consistent gains over SotA methods. authors.elsevier.com/a/1mjUWc7X5IcBX

Bonifacio introduces a conformance-checking framework for pushdown reactive systems based on Visibly Pushdown Languages (VPLs). The accompanying demonstrated its ability to check conformance and generate tests for pushdown reactive system models. doi.org/10.1016/j.scic…

Deshapriya et al. analyzed false positives generated by PMD and SonarQube and examined how their rules align with coding standards. On Stack Overflow and Apache Tomcat, only a few rules generate false positives; ignoring them reduces the rate. doi.org/10.1016/j.scic…

Kekül and Polatgi̇l compared ML models for classifying software quality attributes from user reviews of mobile health apps. The best configuration used TF-IDF vectorization, Random Forest classifier, and SMOTE oversampling, scoring an average F1 of 85%. authors.elsevier.com/a/1michc7X5IcB8