Federica Sarro

Max Hort, Zhenpeng Chen, Jie M. Zhang et al.

This paper provides a comprehensive survey of bias mitigation methods for achieving fairness in Machine Learning (ML) models. We collect a total of 341 publications concerning bias mitigation for ML classifiers. These methods can be distinguished based on their intervention procedure (i.e., pre-processing, in-processing, post-processing) and the technique they apply. We investigate how existing bias mitigation methods are evaluated in the literature. In particular, we consider datasets, metrics and benchmarking. Based on the gathered insights (e.g., What is the most popular fairness metric? How many datasets are used for evaluating bias mitigation methods?), we hope to support practitioners in making informed choices when developing and evaluating new bias mitigation methods.

Saurabhsingh Rajput, Tim Widmayer, Ziyuan Shang et al.

With the increasing usage, scale, and complexity of Deep Learning (DL) models, their rapidly growing energy consumption has become a critical concern. Promoting green development and energy awareness at different granularities is the need of the hour to limit carbon emissions of DL systems. However, the lack of standard and repeatable tools to accurately measure and optimize energy consumption at a fine granularity (e.g., at method level) hinders progress in this area. This paper introduces FECoM (Fine-grained Energy Consumption Meter), a framework for fine-grained DL energy consumption measurement. FECoM enables researchers and developers to profile DL APIs from energy perspective. FECoM addresses the challenges of measuring energy consumption at fine-grained level by using static instrumentation and considering various factors, including computational load and temperature stability. We assess FECoM's capability to measure fine-grained energy consumption for one of the most popular open-source DL frameworks, namely TensorFlow. Using FECoM, we also investigate the impact of parameter size and execution time on energy consumption, enriching our understanding of TensorFlow APIs' energy profiles. Furthermore, we elaborate on the considerations, issues, and challenges that one needs to consider while designing and implementing a fine-grained energy consumption measurement tool. This work will facilitate further advances in DL energy measurement and the development of energy-aware practices for DL systems.

Zhenpeng Chen, Jie M. Zhang, Federica Sarro et al.

Software bias is an increasingly important operational concern for software engineers. We present a large-scale, comprehensive empirical study of 17 representative bias mitigation methods for Machine Learning (ML) classifiers, evaluated with 11 ML performance metrics (e.g., accuracy), 4 fairness metrics, and 20 types of fairness-performance trade-off assessment, applied to 8 widely-adopted software decision tasks. The empirical coverage is much more comprehensive, covering the largest numbers of bias mitigation methods, evaluation metrics, and fairness-performance trade-off measures compared to previous work on this important software property. We find that (1) the bias mitigation methods significantly decrease ML performance in 53% of the studied scenarios (ranging between 42%~66% according to different ML performance metrics); (2) the bias mitigation methods significantly improve fairness measured by the 4 used metrics in 46% of all the scenarios (ranging between 24%~59% according to different fairness metrics); (3) the bias mitigation methods even lead to decrease in both fairness and ML performance in 25% of the scenarios; (4) the effectiveness of the bias mitigation methods depends on tasks, models, the choice of protected attributes, and the set of metrics used to assess fairness and ML performance; (5) there is no bias mitigation method that can achieve the best trade-off in all the scenarios. The best method that we find outperforms other methods in 30% of the scenarios. Researchers and practitioners need to choose the bias mitigation method best suited to their intended application scenario(s).

Rebecca Moussa, Federica Sarro

In the past two decades, several Machine Learning (ML) libraries have become freely available. Many studies have used such libraries to carry out empirical investigations on predictive Software Engineering (SE) tasks. However, the differences stemming from using one library over another have been overlooked, implicitly assuming that using any of these libraries would provide the user with the same or very similar results. This paper aims at raising awareness of the differences incurred when using different ML libraries for software development effort estimation (SEE), one of most widely studied SE prediction tasks. To this end, we investigate 4 deterministic machine learners as provided by 3 of the most popular ML open-source libraries written in different languages (namely, Scikit-Learn, Caret and Weka). We carry out a thorough empirical study comparing the performance of the machine learners on 5 SEE datasets in the two most common SEE scenarios (i.e., out-of-the-box-ml and tuned-ml) as well as an in-depth analysis of the documentation and code of their APIs. The results of our study reveal that the predictions provided by the 3 libraries differ in 95% of the cases on average across a total of 105 cases studied. These differences are significantly large in most cases and yield misestimations of up to approx. 3,000 hours per project. Moreover, our API analysis reveals that these libraries provide the user with different levels of control on the parameters one can manipulate, and a lack of clarity and consistency, overall, which might mislead users. Our findings highlight that the ML library is an important design choice for SEE studies, which can lead to a difference in performance. However, such a difference is under-documented. We conclude by highlighting open-challenges with suggestions for the developers of libraries as well as for the researchers and practitioners using them.

David Williams, Ioakim Avraam, Aldeida Aleti et al.

Automated Program Repair (APR) can reduce the time developers spend debugging, allowing them to focus on other aspects of software development. Automatically generated bug patches are typically validated through software testing. However, this method can lead to patch overfitting, i.e., generating patches that pass the given tests but are still incorrect. Patch correctness assessment (also known as overfitting detection) techniques have been proposed to identify patches that overfit. However, prior work often assessed the effectiveness of these techniques in isolation and on datasets that do not reflect the distribution of correct-to-overfitting patches that would be generated by APR tools in typical use; thus, we still do not know their effectiveness in practice. This work presents the first comprehensive benchmarking study of several patch overfitting detection (POD) methods in a practical scenario. To this end, we curate datasets that reflect realistic assumptions (i.e., patches produced by tools run under the same experimental conditions). Next, we use these data to benchmark six state-of-the-art POD approaches -- spanning static analysis, dynamic testing, and learning-based approaches -- against two baselines based on random sampling (one from prior work and one proposed herein). Our results are striking: Simple random selection outperforms all POD tools for 71% to 96% of cases, depending on the POD tool. This suggests two main takeaways: (1) current POD tools offer limited practical benefit, highlighting the need for novel techniques; (2) any POD tool must be benchmarked on realistic data and against random sampling to prove its practical effectiveness. To this end, we encourage the APR community to continue improving POD techniques and to adopt our proposed methodology for practical benchmarking; we make our data and code available to facilitate such adoption.

Zhaoyang Chu, Jiarui Hu, Xingyu Jiang et al.

We introduce TerminalWorld, a scalable data engine that automatically reverse-engineers high-fidelity evaluation tasks from "in-the-wild" terminal recordings. Processing 80,870 terminal recordings, the engine yields a full benchmark of 1,530 validated tasks, spanning 18 real-world categories, ranging from short everyday operations to workflows exceeding 50 steps, and covering 1,280 unique commands. From these, we curate a Verified subset of 200 representative, manually reviewed tasks. Comprehensive benchmarking on TerminalWorld-Verified across eight frontier models and six agents reveals that current systems still struggle with authentic terminal workflows, achieving a maximum pass rate of only 62.5%. Moreover, TerminalWorld captures real-world terminal capabilities distinct from existing expert-curated benchmarks (e.g., Terminal-Bench), with only a weak correlation to their scores (Pearson r=0.20). The automated engine makes TerminalWorld authentic and scalable by construction, enabling it to evaluate agents in real-world terminal environments as developer practices evolve. Data and code are available at https://github.com/EuniAI/TerminalWorld.

Zhenpeng Chen, Jie M. Zhang, Federica Sarro et al.

Existing research mostly improves the fairness of Machine Learning (ML) software regarding a single protected attribute at a time, but this is unrealistic given that many users have multiple protected attributes. This paper conducts an extensive study of fairness improvement regarding multiple protected attributes, covering 11 state-of-the-art fairness improvement methods. We analyze the effectiveness of these methods with different datasets, metrics, and ML models when considering multiple protected attributes. The results reveal that improving fairness for a single protected attribute can largely decrease fairness regarding unconsidered protected attributes. This decrease is observed in up to 88.3% of scenarios (57.5% on average). More surprisingly, we find little difference in accuracy loss when considering single and multiple protected attributes, indicating that accuracy can be maintained in the multiple-attribute paradigm. However, the effect on F1-score when handling two protected attributes is about twice that of a single attribute. This has important implications for future fairness research: reporting only accuracy as the ML performance metric, which is currently common in the literature, is inadequate.

Minghua Ma, Zhao Tian, Max Hort et al.

Fairness testing aims at mitigating unintended discrimination in the decision-making process of data-driven AI systems. Individual discrimination may occur when an AI model makes different decisions for two distinct individuals who are distinguishable solely according to protected attributes, such as age and race. Such instances reveal biased AI behaviour, and are called Individual Discriminatory Instances (IDIs). In this paper, we propose an approach for the selection of the initial seeds to generate IDIs for fairness testing. Previous studies mainly used random initial seeds to this end. However this phase is crucial, as these seeds are the basis of the follow-up IDIs generation. We dubbed our proposed seed selection approach I&D. It generates a large number of initial IDIs exhibiting a great diversity, aiming at improving the overall performance of fairness testing. Our empirical study reveal that I&D is able to produce a larger number of IDIs with respect to four state-of-the-art seed generation approaches, generating 1.68X more IDIs on average. Moreover, we compare the use of I&D to train machine learning models and find that using I&D reduces the number of remaining IDIs by 29% when compared to the state-of-the-art, thus indicating that I&D is effective for improving model fairness

Emeralda Sesari, Federica Sarro, Ayushi Rastogi

Psychological safety refers to the belief that team members can speak up or make mistakes without fear of negative consequences. While it is recognized as important in traditional software teams, its role in open-source software development remains understudied. Open-source contributors often collaborate without formal roles or structures, where interpersonal relationships can significantly influence participation. Code review, a central and collaborative activity in modern software development, offers a valuable context for observing such team interactions. This paper introduces a framework grounded in psychological safety theory to identify behaviors that signal PS in pull request interactions. We operationalize these behaviors using 10 observable variables derived from 60,684 PRs across 26 popular GitHub repositories and construct a PS index at repository level. We then empirically test the relationship between this index and contributors' short-term (within 1 year) and long-term (over 4-5 years) sustained participation using three logistic regression models. Contributors are more likely to remain active in repositories with higher levels of psychological safety. Psychological safety is positively associated with both short-term and long-term sustained participation. However, prior participation emerges as a stronger predictor of future engagement, reducing the effect of psychological safety when accounted for. This study introduces a a theory-informed framework for measuring psychological safety through pull request data and provides empirical evidence of its relevance in sustaining participation within open-source development.

Xinyue Li, Zhenpeng Chen, Jie M. Zhang et al.

This paper conducts fairness testing of automated pedestrian detection, a crucial but under-explored issue in autonomous driving systems. We evaluate eight state-of-the-art deep learning-based pedestrian detectors across demographic groups on large-scale real-world datasets. To enable thorough fairness testing, we provide extensive annotations for the datasets, resulting in 8,311 images with 16,070 gender labels, 20,115 age labels, and 3,513 skin tone labels. Our findings reveal significant fairness issues, particularly related to age. The proportion of undetected children is 20.14% higher compared to adults. Furthermore, we explore how various driving scenarios affect the fairness of pedestrian detectors. We find that pedestrian detectors demonstrate significant gender biases during night time, potentially exacerbating the prevalent societal issue of female safety concerns during nighttime out. Moreover, we observe that pedestrian detectors can demonstrate both enhanced fairness and superior performance under specific driving conditions, which challenges the fairness-performance trade-off theory widely acknowledged in the fairness literature. We publicly release the code, data, and results to support future research on fairness in autonomous driving.

Alexander E. I. Brownlee, James Callan, Karine Even-Mendoza et al.

Large language models (LLMs) have been successfully applied to software engineering tasks, including program repair. However, their application in search-based techniques such as Genetic Improvement (GI) is still largely unexplored. In this paper, we evaluate the use of LLMs as mutation operators for GI to improve the search process. We expand the Gin Java GI toolkit to call OpenAI's API to generate edits for the JCodec tool. We randomly sample the space of edits using 5 different edit types. We find that the number of patches passing unit tests is up to 75% higher with LLM-based edits than with standard Insert edits. Further, we observe that the patches found with LLMs are generally less diverse compared to standard edits. We ran GI with local search to find runtime improvements. Although many improving patches are found by LLM-enhanced GI, the best improving patch was found by standard GI.

Giovanni Pinna, Jingzhi Gong, David Williams et al.

The rapid adoption of AI-powered coding assistants is transforming software development practices, yet systematic comparisons of their effectiveness across different task types and over time remain limited. This paper presents an empirical study comparing five popular agents (OpenAI Codex, GitHub Copilot, Devin, Cursor, and Claude Code), analyzing 7,156 pull requests (PRs) from the AIDev dataset. Temporal trend analysis reveals heterogeneous evolution patterns: Devin exhibits the only consistent positive trend in acceptance rate (+0.77% per week over 32 weeks), whereas other agents remain largely stable. Our analysis suggests that the PR task type is a dominant factor influencing acceptance rates: documentation tasks achieve 82.1% acceptance compared to 66.1% for new features - a 16 percentage point gap that exceeds typical inter-agent variance for most tasks. OpenAI Codex achieves consistently high acceptance rates across all nine task categories (59.6%-88.6%), with stratified Chi-square tests confirming statistically significant advantages over other agents in several task categories. However, no single agent performs best across all task types: Claude Code leads in documentation (92.3%) and features (72.6%), while Cursor excels in fix tasks (80.4%).

Han Li, Letian Zhu, Bohan Zhang et al.

LLM-based coding agents have shown strong performance on automated issue resolution benchmarks, yet existing evaluations largely focus on final task success, providing limited insight into how agents retrieve and use code context during problem solving. We introduce ContextBench, a process-oriented evaluation of context retrieval in coding agents. ContextBench consists of 1,136 issue-resolution tasks from 66 repositories across eight programming languages, each augmented with human-annotated gold contexts. We further implement an automated evaluation framework that tracks agent trajectories and measures context recall, precision, and efficiency throughout issue resolution. Using ContextBench, we evaluate four frontier LLMs and five coding agents. Our results show that sophisticated agent scaffolding yields only marginal gains in context retrieval ("The Bitter Lesson" of coding agents), LLMs consistently favor recall over precision, and substantial gaps exist between explored and utilized context. ContextBench augments existing end-to-end benchmarks with intermediate gold-context metrics that unbox the issue-resolution process. These contexts offer valuable intermediate signals for guiding LLM reasoning in software tasks.

Giordano d'Aloisio, Sophie Fortz, Carol Hanna et al.

Background: Quantum computing is a rapidly growing new programming paradigm that brings significant changes to the design and implementation of algorithms. Understanding quantum algorithms requires knowledge of physics and mathematics, which can be challenging for software developers. Aims: In this work, we provide a first analysis of how LLMs can support developers' understanding of quantum code. Method: We empirically analyse and compare the quality of explanations provided by three widely adopted LLMs (Gpt3.5, Llama2, and Tinyllama) using two different human-written prompt styles for seven state-of-the-art quantum algorithms. We also analyse how consistent LLM explanations are over multiple rounds and how LLMs can improve existing descriptions of quantum algorithms. Results: Llama2 provides the highest quality explanations from scratch, while Gpt3.5 emerged as the LLM best suited to improve existing explanations. In addition, we show that adding a small amount of context to the prompt significantly improves the quality of explanations. Finally, we observe how explanations are qualitatively and syntactically consistent over multiple rounds. Conclusions: This work highlights promising results, and opens challenges for future research in the field of LLMs for quantum code explanation. Future work includes refining the methods through prompt optimisation and parsing of quantum code explanations, as well as carrying out a systematic assessment of the quality of explanations.

Jingzhi Gong, Sisi Li, Giordano d'Aloisio et al.

Tuning the parameters and prompts for improving AI-based text-to-image generation has remained a substantial yet unaddressed challenge. Hence we introduce GreenStableYolo, which improves the parameters and prompts for Stable Diffusion to both reduce GPU inference time and increase image generation quality using NSGA-II and Yolo. Our experiments show that despite a relatively slight trade-off (18%) in image quality compared to StableYolo (which only considers image quality), GreenStableYolo achieves a substantial reduction in inference time (266% less) and a 526% higher hypervolume, thereby advancing the state-of-the-art for text-to-image generation.

Tosin Fadahunsi, Giordano d'Aloisio, Antinisca Di Marco et al.

Generative models are nowadays widely used to generate graphical content used for multiple purposes, e.g. web, art, advertisement. However, it has been shown that the images generated by these models could reinforce societal biases already existing in specific contexts. In this paper, we focus on understanding if this is the case when one generates images related to various software engineering tasks. In fact, the Software Engineering (SE) community is not immune from gender and ethnicity disparities, which could be amplified by the use of these models. Hence, if used without consciousness, artificially generated images could reinforce these biases in the SE domain. Specifically, we perform an extensive empirical evaluation of the gender and ethnicity bias exposed by three versions of the Stable Diffusion (SD) model (a very popular open-source text-to-image model) - SD 2, SD XL, and SD 3 - towards SE tasks. We obtain 6,720 images by feeding each model with two sets of prompts describing different software-related tasks: one set includes the Software Engineer keyword, and one set does not include any specification of the person performing the task. Next, we evaluate the gender and ethnicity disparities in the generated images. Results show how all models are significantly biased towards male figures when representing software engineers. On the contrary, while SD 2 and SD XL are strongly biased towards White figures, SD 3 is slightly more biased towards Asian figures. Nevertheless, all models significantly under-represent Black and Arab figures, regardless of the prompt style used. The results of our analysis highlight severe concerns about adopting those models to generate content for SE tasks and open the field for future research on bias mitigation in this context.

Giordano d'Aloisio, Tosin Fadahunsi, Jay Choy et al.

Background: Text-to-image generation models are widely used across numerous domains. Among these models, Stable Diffusion (SD) - an open-source text-to-image generation model - has become the most popular, producing over 12 billion images annually. However, the widespread use of these models raises concerns regarding their social and environmental sustainability. Aims: To reduce the harm that SD models may have on society and the environment, we introduce SustainDiffusion, a search-based approach designed to enhance the social and environmental sustainability of SD models. Method: SustainDiffusion searches the optimal combination of hyperparameters and prompt structures that can reduce gender and ethnic bias in generated images while also lowering the energy consumption required for image generation. Importantly, SustainDiffusion maintains image quality comparable to that of the original SD model. Results: We conduct a comprehensive empirical evaluation of SustainDiffusion, testing it against six different baselines using 56 different prompts. Our results demonstrate that SustainDiffusion can reduce gender bias in SD3 by 68%, ethnic bias by 59%, and energy consumption (calculated as the sum of CPU and GPU energy) by 48%. Additionally, the outcomes produced by SustainDiffusion are consistent across multiple runs and can be generalised to various prompts. Conclusions: With SustainDiffusion, we demonstrate how enhancing the social and environmental sustainability of text-to-image generation models is possible without fine-tuning or changing the model's architecture.

Mar Zamorano, Carlos Cetina, Federica Sarro

Video games demand is constantly increasing, which requires the costly production of large amounts of content. Towards this challenge, researchers have developed Search-Based Procedural Content Generation (SBPCG), that is, the (semi-)automated creation of content through search algorithms. We survey the current state of SBPCG, reporting work appeared in the field between 2011-2022 and identifying open research challenges. The results lead to recommendations for practitioners and to the identification of several potential future research avenues for SBPCG.

Zhiwei Fei, Yue Pan, Federica Sarro et al.

Identifying the root cause of a bug remains difficult for many developers because bug reports often lack a bug reproducing test case that reliably triggers the failure. Manually writing such test cases is time-consuming and requires substantial effort to understand the codebase and isolate the failing behavior. To address this challenge, we propose Echo, an agent for generating issue reproducing test cases, which advances previous work in several ways. During generation, Echo strengthens context retrieval by leveraging a code graph and a novel automatic query-refinement strategy. Echo also improves upon previous tools by automatically executing generated test cases, a first-of-its-kind feature that seamlessly integrates into practical development workflows. In addition, Echo generates potential patches and uses the patched version to validate whether a candidate test meets the fail-to-pass criterion and to provide actionable feedback for refinement. Unlike prior bug-reproduction agents that sample and rank multiple candidate tests, Echo generates a single test per issue, offering a better cost-performance trade-off. Experiments on SWT-Bench Verified show that Echo establishes a new state of the art among open-source approaches, achieving a 66.28% success rate.

Vali Tawosi, Afnan Al-Subaihin, Rebecca Moussa et al.

Agile software development is nowadays a widely adopted practise in both open-source and industrial software projects. Agile teams typically heavily rely on issue management tools to document new issues and keep track of outstanding ones, in addition to storing their technical details, effort estimates, assignment to developers, and more. Previous work utilised the historical information stored in issue management systems for various purposes; however, when researchers make their empirical data public, it is usually relevant solely to the study's objective. In this paper, we present a more holistic and versatile dataset containing a wealth of information on more than 500,000 issues from 44 open-source Agile software, making it well-suited to several research avenues, and cross-analyses therein, including effort estimation, issue prioritization, issue assignment and many more. We make this data publicly available on GitHub to facilitate ease of use, maintenance, and extensibility.

Giovani Guizzo, Federica Sarro, Jens Krinke et al.

Mutation testing is an effective approach to evaluate and strengthen software test suites, but its adoption is currently limited by the mutants' execution computational cost. Several strategies have been proposed to reduce this cost (a.k.a. mutation cost reduction strategies), however none of them has proven to be effective for all scenarios since they often need an ad-hoc manual selection and configuration depending on the software under test (SUT). In this paper, we propose a novel multi-objective evolutionary hyper-heuristic approach, dubbed Sentinel, to automate the generation of optimal cost reduction strategies for every new SUT. We evaluate Sentinel by carrying out a thorough empirical study involving 40 releases of 10 open-source real-world software systems and both baseline and state-of-the-art strategies as a benchmark. We execute a total of 4,800 experiments, and evaluate their results with both quality indicators and statistical significance tests, following the most recent best practice in the literature. The results show that strategies generated by Sentinel outperform the baseline strategies in 95% of the cases always with large effect sizes. They also obtain statistically significantly better results than state-of-the-art strategies in 88% of the cases, with large effect sizes for 95% of them. Also, our study reveals that the mutation strategies generated by Sentinel for a given software version can be used without any loss in quality for subsequently developed versions in 95% of the cases. These results show that Sentinel is able to automatically generate mutation strategies that reduce mutation testing cost without affecting its testing effectiveness (i.e. mutation score), thus taking off from the tester's shoulders the burden of manually selecting and configuring strategies for each SUT.

Giordano d'Aloisio, David Williams, Giusy Annunziata et al.

The widespread adoption of Large Language Models (LLMs) raises concerns about the potential harmfulness of their responses. In this paper, we first investigate the harmfulness of responses from four general-purpose LLMs. Next, we propose SafeTune, a multi-objective search-based approach to mitigate harmfulness while increasing response relevance through hyperparameter tuning and system prompt engineering. Our initial evaluation shows that SafeTune significantly reduces the rate of harmful responses generated by Qwen3.5 0.8B and increases prompt-response relevance (both with a large effect size). Among the parameters we explore, we also find that encouraging greater repetition in responses is most impactful in reducing harmfulness while increasing relevance.

Qunying Song, Ali Nouri, Håkan Sivencrona et al.

Autonomous driving systems (ADS) are increasingly deployed in real traffic, yet testing remains fundamentally challenging due to open environments, complex scenarios, and the lack of established processes and metrics. Despite extensive research, a gap persists between academic advances and their applicability in industrial practice. To address this, we conduct an interactive rapid review in collaboration with 21 practitioners from a leading automotive company. Practitioners identified 12 key challenges in ADS testing, and prioritised two as the most critical issues, namely approaches to and completeness of testing for End-to-End (E2E) ADS. We analyzed 17 research studies relevant to these two challenges, most of which focus on generating critical testing scenarios, and subsequently assessed their relevance and applicability in practice. Our study provides the first practitioner-driven review and evaluation of current ADS testing research, reveals practical challenges in ADS testing, offers rapid insights for practitioners, and highlights the need for more context-aware, industry-relevant solutions to bridge the gap between research and practice.

Oorja Majgaonkar, Zhiwei Fei, Xiang Li et al.

The increasing deployment of Large Language Model (LLM) agents for complex software engineering tasks has created a need to understand their problem-solving behaviours beyond simple success metrics. While these agents demonstrate impressive capabilities in automated issue resolution, their decision-making processes remain largely opaque. This paper presents an empirical study of agent trajectories, namely the execution traces capturing the steps agents take when attempting to resolve software issues. We analyse trajectories from three state-of-the-art code agents (OpenHands, SWE-agent, and Prometheus) on the SWE-Bench benchmark, examining both successful and failed attempts. Our investigation reveals several key insights into agent behaviour. First, we identify how distinct problem-solving strategies, such as defensive programming and context gathering, enable success in different scenarios. Second, we find that failed trajectories are consistently longer and exhibit higher variance than successful ones, with failure patterns differing significantly between agents. Third, our fault localisation analysis shows that while most trajectories correctly identify problematic files (72-81\% even in failures), success depends more on achieving approximate rather than exact code modifications. These and other findings unveiled by our study, provide a foundation for understanding agent behaviour through trajectory analysis, contributing to the development of more robust and interpretable autonomous software engineering systems.

Carol Hanna, Karine Even-Mendoza, W. B. Langdon et al.

Despite the operational importance of hot fixes, large-scale evidence on how they reshape routine maintenance workflows, particularly in the era of autonomous coding agents, remains limited. We analyse hot fixes present in over 61,000 GitHub repositories from the Hao-Li/AIDev dataset and find consistent patterns of urgency: reduced collaboration (typically a single contributor), smaller and more targeted changes (median 2-3 commits and files, with <10 line modifications), limited review (often fewer than two reviewers), and substantially fewer test file modifications than regular bug fixes, consistent with their urgency-driven character. Leveraging the same urgency contexts, we examine differences between human- and AI-agent-authored hot fixes, revealing over 10 distinct repair behaviours, thus offering insights into future human-automation collaboration for hot fixing. Our study is the first to empirically analyse hot fix code changes at scale using a repository-level operationalisation of urgency. The comparison of human and agentbehaviours delineates their distinct characteristics, providing a foundation for understanding hot fixing in real-world practice

Jingzhi Gong, Ruizhen Gu, Zhiwei Fei et al.

Agent skills provide modular, task-specific guidance for LLM- based coding agents, but manually tuning skill bundles to balance success rate, cost, and runtime is expensive and fragile. We present SkillMOO, a multi-objective optimization framework that automatically evolves skill bundles using LLM-proposed edits and NSGA-II survivor selection: a solver agent evaluates candidate skill bundles on coding tasks and an optimizer agent proposes bundle edits based on failure analysis. On three SkillsBench software engineering tasks, SkillMOO improves pass rate by up to 131% while reducing cost up to 32% relative to the best baseline per task at low optimization overhead. Pattern analysis reveals pruning and substitution as primary drivers of improvement, suggesting effective bundles favor minimal, focused content over accumulated instructions.

Giordano d'Aloisio, Luca Traini, Federica Sarro et al.

Language models have proven successful across a wide range of software engineering tasks, but their significant computational costs often hinder their practical adoption. To address this challenge, researchers have begun applying various compression strategies to improve the efficiency of language models for code. These strategies aim to optimize inference latency and memory usage, though often at the cost of reduced model effectiveness. However, there is still a significant gap in understanding how these strategies influence the efficiency and effectiveness of language models for code. Here, we empirically investigate the impact of three well-known compression strategies -- knowledge distillation, quantization, and pruning -- across three different classes of software engineering tasks: vulnerability detection, code summarization, and code search. Our findings reveal that the impact of these strategies varies greatly depending on the task and the specific compression method employed. Practitioners and researchers can use these insights to make informed decisions when selecting the most appropriate compression strategy, balancing both efficiency and effectiveness based on their specific needs.

Luís Cruz, João Paulo Fernandes, Maja H. Kirkeby et al.

The environmental impact of Artificial Intelligence (AI)-enabled systems is increasing rapidly, and software engineering plays a critical role in developing sustainable solutions. The "Greening AI with Software Engineering" CECAM-Lorentz workshop (no. 1358, 2025) funded by the Centre Européen de Calcul Atomique et Moléculaire and the Lorentz Center, provided an interdisciplinary forum for 29 participants, from practitioners to academics, to share knowledge, ideas, practices, and current results dedicated to advancing green software and AI research. The workshop was held February 3-7, 2025, in Lausanne, Switzerland. Through keynotes, flash talks, and collaborative discussions, participants identified and prioritized key challenges for the field. These included energy assessment and standardization, benchmarking practices, sustainability-aware architectures, runtime adaptation, empirical methodologies, and education. This report presents a research agenda emerging from the workshop, outlining open research directions and practical recommendations to guide the development of environmentally sustainable AI-enabled systems rooted in software engineering principles.

Zhenpeng Chen, Xinyue Li, Jie M. Zhang et al.

Intersectional fairness is a critical requirement for Machine Learning (ML) software, demanding fairness across subgroups defined by multiple protected attributes. This paper introduces FairHOME, a novel ensemble approach using higher order mutation of inputs to enhance intersectional fairness of ML software during the inference phase. Inspired by social science theories highlighting the benefits of diversity, FairHOME generates mutants representing diverse subgroups for each input instance, thus broadening the array of perspectives to foster a fairer decision-making process. Unlike conventional ensemble methods that combine predictions made by different models, FairHOME combines predictions for the original input and its mutants, all generated by the same ML model, to reach a final decision. Notably, FairHOME is even applicable to deployed ML software as it bypasses the need for training new models. We extensively evaluate FairHOME against seven state-of-the-art fairness improvement methods across 24 decision-making tasks using widely adopted metrics. FairHOME consistently outperforms existing methods across all metrics considered. On average, it enhances intersectional fairness by 47.5%, surpassing the currently best-performing method by 9.6 percentage points.

Maya Larbi, Amal Akli, Mike Papadakis et al.

Large Language Models (LLMs) have demonstrated impressive performance in code generation tasks under idealized conditions, where task descriptions are clear and precise. However, in practice, task descriptions frequently exhibit ambiguity, incompleteness, or internal contradictions. In this paper, we present the first empirical study examining the robustness of state-of-the-art code generation models when faced with such unclear task descriptions. We extend the HumanEval and MBPP benchmarks by systematically introducing realistic task descriptions flaws through guided mutation strategies, producing a dataset that mirrors the messiness of informal developer instructions. We evaluate multiple LLMs of varying sizes and architectures, analyzing their functional correctness and failure modes across task descriptions categories. Our findings reveal that even minor imperfections in task description phrasing can cause significant performance degradation, with contradictory task descriptions resulting in numerous logical errors. Moreover, while larger models tend to be more resilient than smaller variants, they are not immune to the challenges posed by unclear requirements. We further analyze semantic error patterns and identify correlations between description clarity, model behavior, and error types. Our results underscore the critical need for developing LLMs that are not only powerful but also robust to the imperfections inherent in natural user tasks, highlighting important considerations for improving model training strategies, designing more realistic evaluation benchmarks, and ensuring reliable deployment in practical software development environments.

Qunying Song, He Ye, Mark Harman et al.

Autonomous driving systems (ADS) have been an active area of research, with the potential to deliver significant benefits to society. However, before large-scale deployment on public roads, extensive testing is necessary to validate their functionality and safety under diverse driving conditions. Therefore, different testing approaches are required, and achieving effective and efficient testing of ADS remains an open challenge. Recently, generative AI has emerged as a powerful tool across many domains, and it is increasingly being applied to ADS testing due to its ability to interpret context, reason about complex tasks, and generate diverse outputs. To gain a deeper understanding of its role in ADS testing, we systematically analyzed 91 relevant studies and synthesized their findings into six major application categories, primarily centered on scenario-based testing of ADS. We also reviewed their effectiveness and compiled a wide range of datasets, simulators, ADS, metrics, and benchmarks used for evaluation, while identifying 27 limitations. This survey provides an overview and practical insights into the use of generative AI for testing ADS, highlights existing challenges, and outlines directions for future research in this rapidly evolving field.

Serkan Kirbas, Federica Sarro, David Williams

As software in industry grows in size and complexity, so does the volume of engineering data that companies generate and use. Ideally, this data could be used for many purposes, including informing decisions on engineering priorities. However, without a structured representation of the links between different aspects of software development, companies can struggle to identify the root causes of deficiencies or anticipate the effects of changes. In this paper, we report on our experience at Bloomberg in developing a novel tool, dubbed BayesInsights, which provides an interactive interface for visualising causal dependencies across various aspects of the software engineering (SE) process using Bayesian Networks (BNs). We describe our journey from defining network structures using a combination of established literature, expert insight, and structure learning algorithms, to integrating BayesInsights into existing data analytics solutions, and conclude with a mixed-methods evaluation of performance benchmarking and survey responses from 24 senior practitioners at Bloomberg. Our results revealed 95.8% of participants found the tool useful for identifying software delivery challenges at the team and organisational levels, cementing its value as a proof of concept for modelling software delivery and developer experience. BayesInsights is currently in preview, with access granted to seven engineering teams and a wider deployment roadmap in place for the future.

Jingzhi Gong, Yixin Bian, Luis de la Cal et al.

Coding agents powered by LLMs face critical sustainability and scalability challenges in industrial deployment, with single runs consuming over 100k tokens and incurring environmental costs that may exceed optimization benefits. This paper introduces GA4GC, the first framework to systematically optimize coding agent runtime (greener agent) and code performance (greener code) trade-offs by discovering Pareto-optimal agent hyperparameters and prompt templates. Evaluation on the SWE-Perf benchmark demonstrates up to 135x hypervolume improvement, reducing agent runtime by 37.7% while improving correctness. Our findings establish temperature as the most critical hyperparameter, and provide actionable strategies to balance agent sustainability with code optimization effectiveness in industrial deployment.

Rebecca Moussa, Danielle Azar, Federica Sarro

Defect prediction aims at identifying software components that are likely to cause faults before a software is made available to the end-user. To date, this task has been modeled as a two-class classification problem, however its nature also allows it to be formulated as a one-class classification task. Previous studies show that One-Class Support Vector Machine (OCSVM) can outperform two-class classifiers for within-project defect prediction, however it is not effective when employed at a finer granularity (i.e., commit-level defect prediction). In this paper, we further investigate whether learning from one class only is sufficient to produce effective defect prediction model in two other different scenarios (i.e., granularity), namely cross-version and cross-project defect prediction models, as well as replicate the previous work at within-project granularity for completeness. Our empirical results confirm that OCSVM performance remain low at different granularity levels, that is, it is outperformed by the two-class Random Forest (RF) classifier for both cross-version and cross-project defect prediction. While, we cannot conclude that OCSVM is the best classifier, our results still show interesting findings. While OCSVM does not outperform RF, it still achieves performance superior to its two-class counterpart (i.e., SVM) as well as other two-class classifiers studied herein. We also observe that OCSVM is more suitable for both cross-version and cross-project defect prediction, rather than for within-project defect prediction, thus suggesting it performs better with heterogeneous data. We encourage further research on one-class classifiers for defect prediction as these techniques may serve as an alternative when data about defective modules is scarce or not available.

Vali Tawosi, Rebecca Moussa, Federica Sarro

In the last decade, several studies have explored automated techniques to estimate the effort of agile software development. We perform a close replication and extension of a seminal work proposing the use of Deep Learning for Agile Effort Estimation (namely Deep-SE), which has set the state-of-the-art since. Specifically, we replicate three of the original research questions aiming at investigating the effectiveness of Deep-SE for both within-project and cross-project effort estimation. We benchmark Deep-SE against three baselines (i.e., Random, Mean and Median effort estimators) and a previously proposed method to estimate agile software project development effort (dubbed TF/IDF-SVM), as done in the original study. To this end, we use the data from the original study and an additional dataset of 31,960 issues mined from TAWOS, as using more data allows us to strengthen the confidence in the results, and to further mitigate external validity threats. The results of our replication show that Deep-SE outperforms the Median baseline estimator and TF/IDF-SVM in only very few cases with statistical significance (8/42 and 9/32 cases, respectively), thus confounding previous findings on the efficacy of Deep-SE. The two additional RQs revealed that neither augmenting the training set nor pre-training Deep-SE play lead to an improvement of its accuracy and convergence speed. These results suggest that using semantic similarity is not enough to differentiate user stories with respect to their story points; thus, future work has yet to explore and find new techniques and features that obtain accurate agile software development estimates.

Tushar Sharma, Maria Kechagia, Stefanos Georgiou et al.

The advancements in machine learning techniques have encouraged researchers to apply these techniques to a myriad of software engineering tasks that use source code analysis, such as testing and vulnerability detection. Such a large number of studies hinders the community from understanding the current research landscape. This paper aims to summarize the current knowledge in applied machine learning for source code analysis. We review studies belonging to twelve categories of software engineering tasks and corresponding machine learning techniques, tools, and datasets that have been applied to solve them. To do so, we conducted an extensive literature search and identified 479 primary studies published between 2011 and 2021. We summarize our observations and findings with the help of the identified studies. Our findings suggest that the use of machine learning techniques for source code analysis tasks is consistently increasing. We synthesize commonly used steps and the overall workflow for each task and summarize machine learning techniques employed. We identify a comprehensive list of available datasets and tools useable in this context. Finally, the paper discusses perceived challenges in this area, including the availability of standard datasets, reproducibility and replicability, and hardware resources.

Cuiyun Gao, Jichuan Zeng, Federica Sarro et al.

Context: In-app advertising is the primary source of revenue for many mobile apps. The cost of advertising (ad cost) is non-negligible for app developers to ensure a good user experience and continuous profits. Previous studies mainly focus on addressing the hidden performance costs generated by ads, including consumption of memory, CPU, data traffic, and battery. However, there is no research onanalyzing users' perceptions of ads' performance costs to our knowledge. Objective: To fill this gap and better understand the effects of performance costs of in-app ads on user experience, we conduct a study on analyzing user concerns about ads' performance costs. Method: First, we propose RankMiner, an approach to quantify user concerns about specific appissues, including performance costs. Then, based on the usage traces of 20 subject apps, we measure the performance costs of ads. Finally, we conduct correlation analysis on the performance costs and quantified user concerns to explore whether users complain more for higher performance costs. Results: Our findings include the following: (1) RankMiner can quantify users' concerns better than baselines by an improvement of 214% and 2.5% in terms of Pearson correlation coefficient (a metricfor computing correlations between two variables) and NDCG score (a metric for computing accuracyin prioritizing issues), respectively. (2) The performance costs of the with-ads versions are statistically significantly larger than those of no-ads versions with negligible effect size; (3) Users are moreconcerned about the battery costs of ads, and tend to be insensitive to ads' data traffic costs. Conclusion: Our study is complementary to previous work on in-app ads, and can encourage devel-opers to pay more attention to alleviating the most user-concerned performance costs, such as battery cost.

Paul Ralph, Nauman bin Ali, Sebastian Baltes et al.

Empirical Standards are natural-language models of a scientific community's expectations for a specific kind of study (e.g. a questionnaire survey). The ACM SIGSOFT Paper and Peer Review Quality Initiative generated empirical standards for research methods commonly used in software engineering. These living documents, which should be continuously revised to reflect evolving consensus around research best practices, will improve research quality and make peer review more effective, reliable, transparent and fair.

Yixue Zhao, Justin Chen, Adriana Sejfia et al.

UI testing is tedious and time-consuming due to the manual effort required. Recent research has explored opportunities for reusing existing UI tests from an app to automatically generate new tests for other apps. However, the evaluation of such techniques currently remains manual, unscalable, and unreproducible, which can waste effort and impede progress in this emerging area. We introduce FrUITeR, a framework that automatically evaluates UI test reuse in a reproducible way. We apply FrUITeR to existing test-reuse techniques on a uniform benchmark we established, resulting in 11,917 test reuse cases from 20 apps. We report several key findings aimed at improving UI test reuse that are missed by existing work.

Filomena Ferrucci, M-Tahar Kechadi, Pasquale Salza et al.

Genetic Algorithms (GAs) are powerful metaheuristic techniques mostly used in many real-world applications. The sequential execution of GAs requires considerable computational power both in time and resources. Nevertheless, GAs are naturally parallel and accessing a parallel platform such as Cloud is easy and cheap. Apache Hadoop is one of the common services that can be used for parallel applications. However, using Hadoop to develop a parallel version of GAs is not simple without facing its inner workings. Even though some sequential frameworks for GAs already exist, there is no framework supporting the development of GA applications that can be executed in parallel. In this paper is described a framework for parallel GAs on the Hadoop platform, following the paradigm of MapReduce. The main purpose of this framework is to allow the user to focus on the aspects of GA that are specific to the problem to be addressed, being sure that this task is going to be correctly executed on the Cloud with a good performance. The framework has been also exploited to develop an application for Feature Subset Selection problem. A preliminary analysis of the performance of the developed GA application has been performed using three datasets and shown very promising performance.