arXiv subjects

Shiva Nejati

Publications and source records attributed to Shiva Nejati.

At least 19 recordsLinked to original sources

arXiv2026

Synthesizing Behavioural Models of CPS Using Automata Learning and Statistical Machine Learning

Inferring behavioural models from system executions is essential for supporting formal verification and analysis of complex, heterogeneous cyber-physical systems (CPS). Automata learning provides an effective way to infer state machine models from system executions. However, CPS inputs and outputs often consist of numeric time-series data, while automata learning algorithms assume inputs over a finite symbolic alphabet. As a result, raw numeric data must first be abstracted into a finite set of symbols. In this article, we present MELA, a passive automata learning approach enhanced with machine learning to synthesize behavioural models from numeric time-series data generated by CPS. MELA systematically combines statistical machine learning with automata learning to automatically abstract raw numeric signals into interpretable intervals that are strongly correlated with system states. Specifically, MELA uses information-theoretic variable selection and decision-tree-based range abstraction to transform numeric traces into symbolic representations suitable for automata learning. We evaluate MELA on two CPS: a commercial network intrusion detection system developed by our industry partner, RabbitRun Technologies, and a publicly available industrial autopilot benchmark from the aerospace domain. Compared with expertise-based numeric data abstraction, MELA reduces the number of states and transitions in the learned state machines by 49.20% on average, while improving accuracy by 41.71% on average. Furthermore, the learned state machines support system-level requirement verification and help practitioners explore behaviours that are not explicit in the system requirements. We make our implementation and experimental data available online. Keywords: Automata learning, Cyber-physical systems, Behavioural model synthesis, Decision trees, Model checking, Intrusion detection, Simulink.

cs.SE↗

arXiv2026

Question Answering for Diagram-Rich Technical Meeting Videos

Software engineering increasingly relies on asynchronous communication artifacts, including recorded meetings where stakeholders discuss concerns, rationale, and decisions. These meetings often include diagram-based representations of requirements, system behavior, component interactions, and trace dependencies. Accessing knowledge from these meetings is challenging because recordings are long and relevant evidence is distributed across speech, slides, and technical diagrams. This paper reports our industrial experience developing and evaluating LMVQA, an LLM-based multimodal question-answering system for technical meeting videos. Developed in collaboration with engineers at Ciena, LMVQA supports the understanding of requirements and design intent by grounding answers in audio and visual evidence, with explicit handling of diagram-rich content such as requirements and UML diagrams. It processes each video once to build a reusable time-stamped evidence corpus for grounded question answering. Across a Ciena dataset and a public dataset, we show that LMVQA significantly improves answer accuracy compared to a state-of-the-art baseline, from 31% to 94% on the Ciena dataset and from 21% to 88% on the public dataset, with larger gains on diagram-rich videos. We further show that, after one-time indexing, LMVQA reduces average response time from 81.3s to 3.3s on Ciena and from 98.4s to 9.2s on the public dataset, while lowering average token-based LLM API cost by about 75%. Finally, our interviews with three domain experts show that engineers particularly value LMVQA for locating software-engineering-relevant information, revisiting rationale, and tracing answers to specific video segments.

cs.SE↗

arXiv2026

LLM-Based Discovery of Latent Requirements from Stakeholder Conversations: Preliminary Results from Industry

Stakeholder interviews are an important source of information for requirements elicitation, yet many relevant requirements remain implicit in such conversations. Stakeholders frequently describe workflows, challenges, and operational practices without explicitly articulating the software capabilities that could address them. Recent work has considered the use of LLMs to analyze conversational data and extract requirements from stakeholder interviews. Existing approaches, however, primarily focus on identifying explicitly stated requirements, leaving implicit opportunities largely unexplored. In this paper, we present LENS (LLM-Enabled Needs Discovery from Stakeholder Interviews), an approach that analyzes stakeholder interview transcripts to both extract explicit requirements and infer additional latent requirements. LENS performs this inference by reasoning over stakeholder statements together with contextual information about organizational tools and infrastructure. Both extracted and inferred requirements are represented as user stories and linked to transcript excerpts to ensure traceability. We conduct a preliminary evaluation of LENS using twelve stakeholder interview transcripts collected in an industrial setting involving cybersecurity operations. We show that LENS achieves an average F1-score of 84.4% for extracting explicit requirements, while, on average, 75% of the latent requirements identified by LENS were perceived as providing useful automation or time-saving potential by domain experts.

cs.SE↗

arXiv2026

Agentic Persona Generation with Critique-Refinement: An Industrial Evaluation

Personas are widely used in software engineering to support requirements elicitation, design, and validation, but their manual creation is costly, time-consuming, and hard to scale. Recent LLM-based approaches automate persona generation from textual data; however, they typically rely on single-shot generation and subjective evaluations, limiting practical reliability. We present PerGent, an industry-grade method for persona generation built around an iterative critique-refinement loop. Specifically, PerGent uses a generator and a critic LLM agent, coordinated by an orchestrator, to iteratively refine personas using external resources such as interviews, surveys, and job postings through a critique-refinement loop with a user-defined maximum number of rounds. We deploy and evaluate PerGent in an industrial setting at Kinaxis, comparing it with three baselines, including one-shot methods. In an expert in-situ evaluation, PerGent achieved the highest expert approval rate (96.9%), exceeding all baselines. We further compare PerGent-generated personas with best-practice personas manually created by domain experts prior to the adoption of LLMs. Compared to baselines, PerGent reproduces a larger proportion of expert content while also contributing substantial new content beyond the pre-LLM personas. We conclude with lessons learned from deploying and evaluating PerGent at Kinaxis.

cs.SE↗

arXiv2026

Genetic Programming for Self-Adaptive Auto-Scaling of Microservices

Microservice architecture is widely adopted in modern systems, where auto-scaling is critical for satisfying service-level objectives (SLOs). However, determining optimal scaling for microservices is difficult, and reactive resource allocation often leads to costly over- or under-provisioning. We propose AutoSLO, a learning-based, self-adaptive scaling framework that dynamically adjusts microservice replicas to meet SLOs while minimizing resource usage. AutoSLO uses a continuous monitoring-adaptation feedback loop and leverages genetic programming to learn and evolve scaling logic, enabling the deployed microservice system to proactively prevent SLO violations rather than repeatedly searching for one-off scaling actions. We evaluate AutoSLO on two case-study systems -- an online shopping platform and a chatbot based on large language models -- and show that this framework substantially reduces resource usage while maintaining a low frequency of SLO violations, all of which are resolved within a short time window.

cs.SE↗

arXiv2026

Question Answering for Multi-Release Systems: A Case Study at Ciena

Companies regularly have to contend with multi-release systems, where several versions of the same software are in operation simultaneously. Question answering over documents from multi-release systems poses challenges because different releases have distinct yet overlapping documentation. Motivated by the observed inaccuracy of state-of-the-art question-answering techniques on multi-release system documents, we propose QAMR, a chatbot designed to answer questions across multi-release system documentation. QAMR enhances traditional retrieval-augmented generation (RAG) to ensure accuracy in the face of highly similar yet distinct documentation for different releases. It achieves this through a novel combination of pre-processing, query rewriting, and context selection. In addition, QAMR employs a dual-chunking strategy to enable separately tuned chunk sizes for retrieval and answer generation, improving overall question-answering accuracy. We evaluate QAMR using a public software-engineering benchmark as well as a collection of real-world, multi-release system documents from our industry partner, Ciena. Our evaluation yields five main findings: (1) QAMR outperforms a baseline RAG-based chatbot, achieving an average answer correctness of 88.5% and an average retrieval accuracy of 90%, which correspond to improvements of 16.5% and 12%, respectively. (2) An ablation study shows that QAMR's mechanisms for handling multi-release documents directly improve answer accuracy. (3) Compared to its component-ablated variants, QAMR achieves a 19.6% average gain in answer correctness and a 14.0% average gain in retrieval accuracy over the best ablation. (4) QAMR reduces response time by 8% on average relative to the baseline. (5) The automatically computed accuracy metrics used in our evaluation strongly correlate with expert human assessments, validating the reliability of our methodology.

cs.SE↗

arXiv2026

DSL or Code? Evaluating the Quality of LLM-Generated Algebraic Specifications: A Case Study in Optimization at Kinaxis

Model-driven engineering (MDE) provides abstraction and analytical rigour, but industrial adoption in many domains has been limited by the cost of developing and maintaining models. Large language models (LLMs) can help shift this cost balance by supporting direct generation of models from natural-language (NL) descriptions. For domain-specific languages (DSLs), however, LLM-generated models may be less accurate than LLM-generated code in mainstream languages such as Python, due to the latter's dominance in LLM training corpora. We investigate this issue in mathematical optimization, with AMPL, a DSL with established industrial use. We introduce EXEOS, an LLM-based approach that derives AMPL models and Python code from NL problem descriptions and iteratively refines them with solver feedback. Using a public optimization dataset and real-world supply-chain cases from our industrial partner Kinaxis, we evaluate generated AMPL models against Python code in terms of executability and correctness. An ablation study with two LLM families shows that AMPL is competitive with, and sometimes better than, Python, and that our design choices in EXEOS improve the quality of generated specifications.

cs.SE↗

arXiv2025

The Impact of Critique on LLM-Based Model Generation from Natural Language: The Case of Activity Diagrams

Large Language Models (LLMs) show strong potential for automating model generation from natural-language descriptions. A common approach begins with an initial model generation, followed by an iterative critique-refine loop in which the model is evaluated for issues and refined based on those issues. This process needs to address: (1) structural correctness -- compliance with well-formedness rules -- and (2) semantic alignment -- accurate reflection of the intended meaning in the source text. We present LADEX (LLM-based Activity Diagram Extractor), a pipeline for deriving activity diagrams from natural-language process descriptions using an LLM-driven critique-refine process. Structural checks in LADEX can be performed either algorithmically or by an LLM, while alignment checks are performed by an LLM. We design five ablated variants of LADEX to study: (i) the impact of the critique-refine loop itself, (ii) the role of LLM-based semantic checks, and (iii) the comparative effectiveness of algorithmic versus LLM-based structural checks. To evaluate LADEX, we compare generated diagrams with expert ground truths using a trace-based behavioural and an LLM-based matcher. This enables automated measurement of correctness (whether the generated activity diagram includes the ground-truth nodes) and completeness (how many of the ground-truth nodes the generated activity diagram covers). Experiments on two datasets -- a public-domain dataset and an industry dataset from our collaborator, Ciena -- indicate: (1) Both matchers yield similar completeness and correctness comparisons. (2) The critique-refine loop improves structural validity, correctness, and completeness compared to single-pass generation. (3) Activity diagrams refined based on algorithmic structural checks achieve structural consistency, whereas those refined based on LLM-based checks often still show structural inconsistencies.

cs.SE↗

arXiv2025

Automated Test Validators for Flaky Cyber-Physical System Simulators: Approach and Evaluation

Simulation-based testing of cyber-physical systems (CPS) is costly due to the time-consuming execution of CPS simulators. In addition, CPS simulators may be flaky, leading to inconsistent test outcomes and requiring repeated test re-execution for reliable test verdicts. Many test inputs within the input space of CPS may not effectively exercise the behaviour of the system under test (SUT) -- for instance, those that violate system preconditions, exceed operational design domain (ODD) limits, or represent inherently safe scenarios. In this article, we propose to use test validators to filter out such test inputs before execution. We describe two methods for generating test validators: one using genetic programming (GP) that employs well-known spectrum-based fault localization (SBFL) ranking formulas, namely Ochiai, Tarantula, and Naish, as fitness functions; and the other using decision trees (DT) and decision rules (DR). We evaluate our test validators through case studies in the domains of aerospace, networking and autonomous driving. We show that test validators generated using GP with Ochiai are significantly more accurate than those generated using GP with Tarantula and Naish or using DT or DR. Moreover, this accuracy advantage remains even when accounting for the flakiness of the simulator. We further show that our test validators generated by GP with Ochiai are robust against flakiness with only 4% average variation in their accuracy results across four different network and autonomous-driving systems with flaky behaviours. Finally, we show that, on average, 88.7% of the assertions inferred by our approach align or overlap with requirements precondition violations, ODD-limit violations, and nominal safe conditions extracted from technical standards and empirical results in the literature.

cs.SE↗

arXiv2025

Effort-Optimized, Accuracy-Driven Labelling and Validation of Test Inputs for DL Systems: A Mixed-Integer Linear Programming Approach

Software systems increasingly include AI components based on deep learning (DL). Reliable testing of such systems requires near-perfect test-input validity and label accuracy, with minimal human effort. Yet, the DL community has largely overlooked the need to build highly accurate datasets with minimal effort, since DL training is generally tolerant of labelling errors. This challenge, instead, reflects concerns more familiar to software engineering, where a central goal is to construct high-accuracy test inputs, with accuracy as close to 100% as possible, while keeping associated costs in check. In this article we introduce OPAL, a human-assisted labelling method that can be configured to target a desired accuracy level while minimizing the manual effort required for labelling. The main contribution of OPAL is a mixed-integer linear programming (MILP) formulation that minimizes labelling effort subject to a specified accuracy target. To evaluate OPAL we instantiate it for two tasks in the context of testing vision systems: automatic labelling of test inputs and automated validation of test inputs. Our evaluation, based on more than 2500 experiments performed on nine datasets, comparing OPAL with eight baseline methods, shows that OPAL, relying on its MILP formulation, achieves an average accuracy of 98.8%, while cutting manual labelling by more than half. OPAL significantly outperforms automated labelling baselines in labelling accuracy across all nine datasets, when all methods are provided with the same manual-labelling budget. For automated test-input validation, on average, OPAL reduces manual effort by 28.8% while achieving 4.5% higher accuracy than the SOTA test-input validation baselines. Finally, we show that augmenting OPAL with an active-learning loop leads to an additional 4.5% reduction in required manual labelling, without compromising accuracy.

cs.CV↗

arXiv2025

Simulink Mutation Testing using CodeBERT

We present BERTiMuS, an approach that uses CodeBERT to generate mutants for Simulink models. BERTiMuS converts Simulink models into textual representations, masks tokens from the derived text, and uses CodeBERT to predict the masked tokens. Simulink mutants are obtained by replacing the masked tokens with predictions from CodeBERT. We evaluate BERTiMuS using Simulink models from an industrial benchmark, and compare it with FIM -- a state-of-the-art mutation tool for Simulink. We show that, relying exclusively on CodeBERT, BERTiMuS can generate the block-based Simulink mutation patterns documented in the literature. Further, our results indicate that: (a) BERTiMuS is complementary to FIM, and (b) when one considers a requirements-aware notion of mutation testing, BERTiMuS outperforms FIM.

cs.SE↗

arXiv2025

Test Input Validation for Vision-based DL Systems: An Active Learning Approach

Testing deep learning (DL) systems requires extensive and diverse, yet valid, test inputs. While synthetic test input generation methods, such as metamorphic testing, are widely used for DL testing, they risk introducing invalid inputs that do not accurately reflect real-world scenarios. Invalid test inputs can lead to misleading results. Hence, there is a need for automated validation of test inputs to ensure effective assessment of DL systems. In this paper, we propose a test input validation approach for vision-based DL systems. Our approach uses active learning to balance the trade-off between accuracy and the manual effort required for test input validation. Further, by employing multiple image-comparison metrics, it achieves better results in classifying valid and invalid test inputs compared to methods that rely on single metrics. We evaluate our approach using an industrial and a public-domain dataset. Our evaluation shows that our multi-metric, active learning-based approach produces several optimal accuracy-effort trade-offs, including those deemed practical and desirable by our industry partner. Furthermore, provided with the same level of manual effort, our approach is significantly more accurate than two state-of-the-art test input validation methods, achieving an average accuracy of 97%. Specifically, the use of multiple metrics, rather than a single metric, results in an average improvement of at least 5.4% in overall accuracy compared to the state-of-the-art baselines. Incorporating an active learning loop for test input validation yields an additional 7.5% improvement in average accuracy, bringing the overall average improvement of our approach to at least 12.9% compared to the baselines.

cs.SE↗

arXiv2024

Can Search-Based Testing with Pareto Optimization Effectively Cover Failure-Revealing Test Inputs?

Search-based software testing (SBST) is a widely adopted technique for testing complex systems with large input spaces, such as Deep Learning-enabled (DL-enabled) systems. Many SBST techniques focus on Pareto-based optimization, where multiple objectives are optimized in parallel to reveal failures. However, it is important to ensure that identified failures are spread throughout the entire failure-inducing area of a search domain and not clustered in a sub-region. This ensures that identified failures are semantically diverse and reveal a wide range of underlying causes. In this paper, we present a theoretical argument explaining why testing based on Pareto optimization is inadequate for covering failure-inducing areas within a search domain. We support our argument with empirical results obtained by applying two widely used types of Pareto-based optimization techniques, namely NSGA-II (an evolutionary algorithm) and OMOPSO (a swarm-based Pareto-optimization algorithm), to two DL-enabled systems: an industrial Automated Valet Parking (AVP) system and a system for classifying handwritten digits. We measure the coverage of failure-revealing test inputs in the input space using a metric that we refer to as the Coverage Inverted Distance quality indicator. Our results show that NSGA-II-based search and OMOPSO are not more effective than a na\"ive random search baseline in covering test inputs that reveal failures. The replication package for this study is available in a GitHub repository.

cs.SE↗

arXiv2024

Bridging the Gap between Real-world and Synthetic Images for Testing Autonomous Driving Systems

Deep Neural Networks (DNNs) for Autonomous Driving Systems (ADS) are typically trained on real-world images and tested using synthetic simulator images. This approach results in training and test datasets with dissimilar distributions, which can potentially lead to erroneously decreased test accuracy. To address this issue, the literature suggests applying domain-to-domain translators to test datasets to bring them closer to the training datasets. However, translating images used for testing may unpredictably affect the reliability, effectiveness and efficiency of the testing process. Hence, this paper investigates the following questions in the context of ADS: Could translators reduce the effectiveness of images used for ADS-DNN testing and their ability to reveal faults in ADS-DNNs? Can translators result in excessive time overhead during simulation-based testing? To address these questions, we consider three domain-to-domain translators: CycleGAN and neural style transfer, from the literature, and SAEVAE, our proposed translator. Our results for two critical ADS tasks -- lane keeping and object detection -- indicate that translators significantly narrow the gap in ADS test accuracy caused by distribution dissimilarities between training and test data, with SAEVAE outperforming the other two translators. We show that, based on the recent diversity, coverage, and fault-revealing ability metrics for testing deep-learning systems, translators do not compromise the diversity and the coverage of test data, nor do they lead to revealing fewer faults in ADS-DNNs. Further, among the translators considered, SAEVAE incurs a negligible overhead in simulation time and can be efficiently integrated into simulation-based testing. Finally, we show that translators increase the correlation between offline and simulation-based testing results, which can help reduce the cost of simulation-based testing.

cs.SE↗

arXiv2024

Developing a Llama-Based Chatbot for CI/CD Question Answering: A Case Study at Ericsson

This paper presents our experience developing a Llama-based chatbot for question answering about continuous integration and continuous delivery (CI/CD) at Ericsson, a multinational telecommunications company. Our chatbot is designed to handle the specificities of CI/CD documents at Ericsson, employing a retrieval-augmented generation (RAG) model to enhance accuracy and relevance. Our empirical evaluation of the chatbot on industrial CI/CD-related questions indicates that an ensemble retriever, combining BM25 and embedding retrievers, yields the best performance. When evaluated against a ground truth of 72 CI/CD questions and answers at Ericsson, our most accurate chatbot configuration provides fully correct answers for 61.11% of the questions, partially correct answers for 26.39%, and incorrect answers for 12.50%. Through an error analysis of the partially correct and incorrect answers, we discuss the underlying causes of inaccuracies and provide insights for further refinement. We also reflect on lessons learned and suggest future directions for further improving our chatbot's accuracy.

cs.SE↗

arXiv2024

Enhancing Automata Learning with Statistical Machine Learning: A Network Security Case Study

Intrusion detection systems are crucial for network security. Verification of these systems is complicated by various factors, including the heterogeneity of network platforms and the continuously changing landscape of cyber threats. In this paper, we use automata learning to derive state machines from network-traffic data with the objective of supporting behavioural verification of intrusion detection systems. The most innovative aspect of our work is addressing the inability to directly apply existing automata learning techniques to network-traffic data due to the numeric nature of such data. Specifically, we use interpretable machine learning (ML) to partition numeric ranges into intervals that strongly correlate with a system's decisions regarding intrusion detection. These intervals are subsequently used to abstract numeric ranges before automata learning. We apply our ML-enhanced automata learning approach to a commercial network intrusion detection system developed by our industry partner, RabbitRun Technologies. Our approach results in an average 67.5% reduction in the number of states and transitions of the learned state machines, while achieving an average 28% improvement in accuracy compared to using expertise-based numeric data abstraction. Furthermore, the resulting state machines help practitioners in verifying system-level security requirements and exploring previously unknown system behaviours through model checking and temporal query checking. We make our implementation and experimental data available online.

cs.CR↗

arXiv2024

Requirements-driven Slicing of Simulink Models Using LLMs

Model slicing is a useful technique for identifying a subset of a larger model that is relevant to fulfilling a given requirement. Notable applications of slicing include reducing inspection effort when checking design adequacy to meet requirements of interest and when conducting change impact analysis. In this paper, we present a method based on large language models (LLMs) for extracting model slices from graphical Simulink models. Our approach converts a Simulink model into a textual representation, uses an LLM to identify the necessary Simulink blocks for satisfying a specific requirement, and constructs a sound model slice that incorporates the blocks identified by the LLM. We explore how different levels of granularity (verbosity) in transforming Simulink models into textual representations, as well as the strategy used to prompt the LLM, impact the accuracy of the generated slices. Our preliminary findings suggest that prompts created by textual representations that retain the syntax and semantics of Simulink blocks while omitting visual rendering information of Simulink models yield the most accurate slices. Furthermore, the chain-of-thought and zero-shot prompting strategies result in the largest number of accurate model slices produced by our approach.

cs.SE↗

arXiv2024

Generating Minimalist Adversarial Perturbations to Test Object-Detection Models: An Adaptive Multi-Metric Evolutionary Search Approach

Deep Learning (DL) models excel in computer vision tasks but can be susceptible to adversarial examples. This paper introduces Triple-Metric EvoAttack (TM-EVO), an efficient algorithm for evaluating the robustness of object-detection DL models against adversarial attacks. TM-EVO utilizes a multi-metric fitness function to guide an evolutionary search efficiently in creating effective adversarial test inputs with minimal perturbations. We evaluate TM-EVO on widely-used object-detection DL models, DETR and Faster R-CNN, and open-source datasets, COCO and KITTI. Our findings reveal that TM-EVO outperforms the state-of-the-art EvoAttack baseline, leading to adversarial tests with less noise while maintaining efficiency.

cs.SE↗