SUMMARY & CONCLUSIONSThis paper presents a novel process for validating and adapting design models (e.g., Simulink models) for use in system safety assessments. The core of the methodology lies in the orchestrated use of Large Language Models (LLMs) and algorithmic investigations to bridge the gap between textual specification documents and functional Simulink models. The process was developed by analyzing the information available in various specification documents and Simulink models from different stages of development. It focuses on deriving and comparing architectural components, connections, and implemented behavior to assess a model's suitability for safety analysis. Once validated, the model is adapted to incorporate safety requirements, such as failure conditions, enabling automated analysis via model checking. The process was prototypically implemented and evaluated on an example landing gear system. It demonstrates the applicability of state-of-the-art LLMs for model validation tasks.

Code review comment generation suggests natural-language feedback for code changes. Although recent work increasingly focuses on generative large language models, retrieval methods remain important, especially as components of retrieval-augmented generation systems. Existing approaches predominantly retrieve comments indirectly through code similarity. We introduce CReaTE+, which instead performs direct cross-modal retrieval in the shared, code review-aware embedding space of our prior CReaTE model: query code changes are matched directly against review comment embeddings. We evaluate retrieval mode and domain-specific fine-tuning across LAURA-DS, CRER-DS, and TUFANO-DS using lexical and semantic metrics. Results show that our domain-specific CReaTE family is strongest overall: CReaTE+ performs best on LAURA-DS and CRER-DS, while the original CReaTE model remains best on TUFANO-DS. Thus, while cross-modal retrieval combined with code review-specific fine-tuning is the most effective strategy, dataset properties can also favor code-similarity-based retrieval.

Path planning for high degrees-of-freedom robots in a collaborative workspace requires real-time adaptability to continuous changes in the environment. Intelligent path adaptations that account for the motion characteristics and attributes of objects in the robot's workspace are essential for a closer human-robot collaboration. This paper presents a reinforcement learning (RL)-based path planning approach that identifies and intelligently adapts the robot's path to obstacles based on their motion patterns of varying dimensionality ranging from one-dimensional linear to three-dimensional helical obstacle trajectories. The experiments indicate that the RL algorithm learned to proactively redirect the robot trajectories to regions of reduced collision risk based on the obstacle motion, resulting in higher success rates than conventional planners, such as artificial potential fields. The planner was tested against repeated high-speed (linear speeds up to 2.5 m/s and angular velocities up to 8p rad/s) path obstructions by dynamic obstacles with noisy, perturbed trajectories. The results highlight the adaptive potential of RL-based path planning for next-generation cobot applications in human-robot collaboration.

Modern Code Review (MCR) is a standard practice in software engineering, yet it demands substantial time and resource investments. Recent research has increasingly explored automating core review tasks using machine learning (ML) and deep learning (DL). As a result, there is substantial variability in task definitions, datasets, and evaluation procedures. This study provides the first comprehensive analysis of MCR automation research, aiming to characterize the field's evolution, formalize learning tasks, highlight methodological challenges, and offer actionable recommendations to guide future research. Focusing on the primary code review tasks, we systematically surveyed 691 publications and identified 24 relevant studies published between May 2015 and April 2024. Each study was analyzed in terms of tasks, models, metrics, baselines, results, validity concerns, and artifact availability. In particular, our analysis reveals significant potential for standardization, including 48 task metric combinations, 22 of which were unique to their original paper, and limited dataset reuse. We highlight challenges and derive concrete recommendations for examples such as the temporal bias threat, which are rarely addressed so far. Our work contributes to a clearer overview of the field, supports the framing of new research, helps to avoid pitfalls, and promotes greater standardization in evaluation practices.

The use of large language models like ChatGPT in code review offers promising efficiency gains but also raises concerns about correctness and safety. Existing evaluation methods for code review generation either rely on automatic comparisons to a single ground truth, which fails to capture the variability of human perspectives, or on subjective assessments of "usefulness", a highly ambiguous concept. We propose a novel evaluation approach based on what we call multi-subjective ranking. Using a dataset of 280 self-contained code review requests and corresponding comments from CodeReview StackExchange, multiple human judges ranked the quality of ChatGPT-generated comments alongside the top human responses from the platform. Results show that ChatGPT's comments were ranked significantly better than human ones, even surpassing StackExchange's accepted answers. Going further, our proposed method motivates and enables more meaningful assessments of generative AI's performance in code review, while also raising awareness of potential risks of unchecked integration into review processes.

In conclusion, this paper aims to provide a solution for bridging the gap between static and dynamic system design representations while highlighting the importance of model reuse for cross-functional collaboration, particularly in the domain of functional safety. By ensuring model continuity and enabling the reuse of model artifacts for automated safety assessment, our approach contributes to a more rigorous and efficient system design and analysis process, ultimately improving the overall safety and reliability of complex systems. This paper presents a model continuity approach and supporting model transformations to bridge the gap between static and dynamic system design representations. Our approach employs a semantics-aware transformation of a system's static architecture into the structural foundation of a dynamic system model. This transformation encompasses: Hierarchical Decompositions: Translating functional, logical, and technical decompositions with their relationships. Failure Injection Points: Mapping defined failure injection points to the dynamic model. Design Retransformation: Enabling updates to the static architecture based on adjustments made during dynamic modeling. When enriched with behavioral details, we demonstrate that the transformed model is a robust starting point for automated formal safety analysis. In a case study from the aerospace domain, we employ model-checking techniques to generate minimal failure mode cut sets initially defined within the static structural architecture. The minimal failure mode cut sets should replace the often-used fault trees.

The assurance of functional safety in complex, softwareintensive systems is a critical challenge across numerous industries. This paper examines the integration of MBSA within Functional safety standards, with a specific focus on the aerospace domain through SAE Aerospace Recommended Practice (ARP) 4761A. ARP4761A is notable for being one of the first comprehensive industry standards to incorporate detailed MBSA guidance formally. This paper analyzes this MBSA methodology, extracting its core process steps, modeling elements (like the Failure Propagation Model - FPM), and algorithmic output generation methods. Furthermore, the paper explores the requirements for the support of computational tools, focusing on the capabilities of model verification techniques to verify safety properties and compute safety metrics within the MBSA framework. A comparative overview of selected MBSA tools and ecosystems against these derived requirements is also presented. The insights aim to contribute to a broader understanding of MBSA application within standardized functional safety practices, using aerospace as a leading example.

Statistical Model Checking (SMC) becomes imprac-tical for hazards with probabilities as small as 10−9. Importance Sampling (IS) can help, but its effectiveness hinges on a good sampling distribution. We present a symbolic-simulation method that constructs such a distribution for Discrete-Time Markov Chains (DTMCs) used in model-based safety assessment. First, we partition all first-passage executions to the hazard into Path-Influence Regions (PIRs), repair-aware classes induced by minimal critical failure mode sets. For each PIR, we encode cut sequences with Bounded Model Checking (BMC) and use Opti-mization Modulo Theories (OMT) to synthesize two probabilistic extremal witnesses (most and least likely). We then solve a convex fitting problem over locally normalized IS weights to equalize (or tightly bound) the likelihood ratios on these extremals, which by affinity controls the entire PIR and is sufficient for computing the optimal distribution. The resulting estimator is unbiased and, under mild conditions, enjoys Bounded Relative Error (BRE) for which we provide a proof sketch and rare-event preliminaries. In experiments on nine benchmarks, our method improves accuracy, reduces relative error, and increases robustness compared to IS baselines (PLASMA and PathZVA), especially for large models and long horizons.

Machine learning models typically perform reliably only on inputs drawn from the distribution they were trained on, making Out-of-Distribution (OOD) detection essential for safety-critical applications. While exposing models to example outliers during training is one of the most effective ways to enhance OOD detection, recent studies suggest that synthetically generated outliers can also act as regularizers for deep neural networks. In this paper, we propose an augmentation scheme for synthetic outliers that regularizes a classifier's energy function by adversarially lowering the outliers' energy during training. We demonstrate that our method improves OOD detection performance and adversarial robustness on OOD data on several image classification benchmarks. Additionally, we show that our approach preserves in-distribution generalization. Our code is publicly available.11https://github.com/2mey10/Adversarial-OE

Code review automation is an important area of research given the time and effort required by traditional processes. Evaluating comment generation models requires automated metrics that quantify the similarity between generated and human-authored comments. Prior work has relied on lexical metrics like BLEU and ExactMatches, which fail to capture semantic differences when phrasing varies. Sentence embedding models offer a promising alternative; however, so far there has been no evidence of their superiority in capturing semantic similarity on code review data. In an empirical study with human subjects, we use OpenAI embeddings to assess semantic similarity in review comments. Our analyses show that the embedding-based metric aligns more closely with human annotations than BLEU. Furthermore, our method to support metric choices with empirical evidence may represent a significant advancement in the domain.

Out-of-distribution (OOD) detection is essential for ensuring the reliability of deep learning models operating in open-world scenarios. Current OOD detectors mainly rely on statistical models to identify unusual patterns in the latent representations of a deep neural network. This work proposes to augment existing OOD detectors with probabilistic reasoning, utilizing Markov logic networks (MLNs). MLNs connect first-order logic with probabilistic reasoning to assign probabilities to inputs based on weighted logical constraints defined over human-understandable concepts, which offers improved explainability. Through extensive experiments on multiple datasets, we demonstrate that MLNs can significantly enhance the performance of a wide range of existing OOD detectors while maintaining computational efficiency. Furthermore, we introduce a simple algorithm for learning logical constraints for OOD detection from a dataset and showcase its effectiveness.

Modern software development relies on the reuse of code via Application Programming Interfaces (APIs). Such reuse relieves developers from learning and developing established algorithms and data structures anew, enabling them to focus on their problem at hand. However, there is also the risk of misusing an API due to a lack of understanding or proper documentation. While many techniques target API misuse detection, only limited efforts have been put into automatically repairing API misuses. In this paper, we present our advances on our technique API-Specific Automated Program Repair (ASAP-Repair). ASAP-Repair is intended to fix API misuses based on API Usage Graphs (AUGs) by leveraging API usage templates of state-of-the-art API misuse detectors. We demonstrate that ASAP-Repair is in principle applicable on an established API misuse dataset. Moreover, we discuss next steps and challenges to evolve ASAP-Repair towards a full-fledged Automatic Program Repair (APR) technique.

Multi-spectral satellite images of the Earth’s surface are used in various applications, from water quality assessment and urban planning to climate monitoring, disaster response, infrastructure oversight, and agricultural surveillance. Many of these applications would benefit from higher spatial and temporal resolution of observations, which could be achieved by combining observations from several sources. This study introduces a deep learning-based pipeline to harmonize the spectral and spatial discrepancies between the Landsat-8 and Sentinel-2 Earth Observation satellites. Through established image quality metrics, we demonstrate a significant enhancement in the spatial resolution of Landsat-8 images. Field observation experiments show that leveraging unified images from both satellites increases the availability of cloud-free images by 21% annually on average in our study area. Additionally, our pipeline enhances the Normalized Difference Vegetation Index (NDVI) correlation between Landsat-8 and Sentinel-2 observations by about 4.9%, offering significant performance gains in a downstream crop segmentation task. Our 100M parameter model, trained on European data, generalizes to most regions with only minor limitations. Furthermore, we show that the pipeline can provide uncertainty estimates for its outputs, which are valuable for decision-making in downstream applications. Our code base is available online: https://github.com/venkatesh-thiru/Deep-Harmonization.

Machine learning models often only generalize reliably to samples from their training distribution which motivates out-of-distribution (OOD) detection in safety-critical applications. Current OOD detection methods, however, tend to be domain agnostic and are incapable of incorporating prior knowledge about the structure of the training distribution. To address this limitation, we introduce a novel, neuro-symbolic OOD detection algorithm that combines a deep learning-based perception system with a first-order logic-based knowledge representation. A reasoning system uses this knowledge base at run-time to infer whether inputs are consistent with prior knowledge about the training distribution. This not only enhances performance but also fosters a level of explainability that is particularly beneficial in safety-critical contexts.

Machine Learning models often only generalize reliably to samples from the training distribution. Consequentially, detecting when input data is out-of-distribution (OOD) is crucial, especially in safety-critical applications. Current OOD detection methods, however, tend to be domain agnostic and often fail to incorporate valuable prior knowledge about the structure of the training distribution. To address this limitation, we introduce a novel, hybrid OOD detection algorithm that combines a deep learning-based perception system with a first-order logic-based knowledge representation. A logical reasoning system uses this knowledge base at run-time to infer whether inputs are consistent with prior knowledge about the training distribution. In contrast to purely neural systems, the structured knowledge representation allows humans to inspect and modify the rules that govern the OOD detectors’ behavior. This not only enhances performance but also fosters a level of explainability that is particularly beneficial in safety-critical contexts. We demonstrate the effectiveness of our method through experiments on several datasets and discuss advantages and limitations. Our code is available online. 1

Deep neural networks (DNNs) are prone to making wrong predictions with high confidence for data that does not stem from their training distribution. Consequentially, out-of-distribution (OOD) detection is important in safety-critical applications, as it identifies such inputs. Using prior knowledge about the training distribution through formal constraints has shown promise in enhancing OOD detection. However, developing and maintaining formal knowledge bases can be cumbersome. Large language models (LLMs) have recently excelled in various natural language processing tasks. In this study, we investigate the use of LLMs for OOD detection, where domain constraints are expressed in natural language. Our results indicate that LLMs can outperform random guessing by leveraging general world knowledge learned during training. Moreover, LLMs can par with methods based on formal constraints when supplemented with domain-specific constraints articulated in natural language.

The request for computer-aided system verification approaches increases with rising system complexity.So, integrating formal verification approaches, e.g., model checking, into the typical engineering workflow could help keep up with this rising complexity.Therefore, however, such an analysis of a system model must cover widely applied design specification languages, e.g., Matlab Simulink or Modelica.This paper shall provide an approach for transforming a Simulink model into the input language of a model checking verification tool.Therefore, we modeled a widely applied case study of an aircraft landing gear system to provide a formal framework that can later be translated into a Kripke structure.A widely used formalism for verification tool input languages.The set of Matlab Simulink elements for which we provide a translation is derived from the model itself.To prove that the translation preserves the model's semantics, we also define a formal representation of the modeled Simulink elements.The translation enables us to apply several model checking tools for formal verification.

This companion paper focuses on challenging aspects of reproducibility that emerge in anomaly detection with Deep Neural Networks. We provide motivating examples based on our work and present mitigation strategies. Furthermore, we document a trade-off between the complexity of experiments and the strength of the empirical evidence obtained through them, both of which impact different types of reproducibility. Ultimately, we argue that the reproducibility of inferences should be prioritized over the reproducibility of exact numerical results.

The performance of machine learning models can be significantly impacted by variations in data quality. Typically, conventional model testing does not examine how robust the model would be in the face of potential data quality deterioration. In an industrial use case, however, data quality is a pertinent issue, as sensors are susceptible to a variety of technical and external issues that may result in poor data quality over time. In order to develop robust machine learning models, industrial data scientists must understand the sensitivity of their models against data quality issues, through the application of an appropriate and comprehensive testing solution. In this work, we propose a generic framework for systematically analyzing the impact of data quality issues on the performance of machine learning models by intentionally applying gradual perturbations to the original time series data. The evaluation is performed using a benchmark industrial process consisting of multivariate time series from sensors in a complex chemical process.