We present Flow-Induced Diagonal Gaussian Processes (FiD-GP), a compression framework that incorporates a compact inducing weight matrix to project a neural network’s weight uncertainty into a lower-dimensional subspace. Critically, FiD-GP relies on normalising flow variational posterior and spectral regularisations to augment its expressiveness and align the inducing subspace with feature-gradient geometry through a numerically stable projection mechanism objective. Furthermore, we demonstrate how the prediction framework in FiD-GP can help to design a single pass projection for Out-of-Distribution (OoD) detection. Our analysis shows that FiD-GP improves uncertainty estimation ability on various tasks compared with SVGP-based baselines, satisfies tight spectral residual bounds with theoretically guaranteed OoD detection, and significantly compresses the neural network’s storage requirements at the cost of increased inference computation dependent on the number of inducing weights employed. Specifically, in a comprehensive empirical study spanning regression, image classification, semantic segmentation, and Out-of-Distribution detection benchmarks, it significantly cuts Bayesian training cost, compresses parameters by roughly 51%, reduces model size by about 75%, and matches state-of-the-art accuracy and uncertainty estimation.

Context: Developer eXperience (Dev-X) focuses on ensuring a better experience for software developers, while still achieving development goals. Indeed, several existing studies suggest that improved Dev-X may actually result in improved development productivity. But a comprehensive understanding of how to empirically assess Dev-X remains limited. Aims and Method: This paper determines the empirical coverage of Dev-X facets to date by analyzing employed empirical designs and identifying quality issues or guidelines for assessing Dev-X. To this end, it presents a mapping study of 231 articles that employed empirical methods to explore the Dev-X landscape. Results: It identifies 16 facets of Dev-X —spanning emotions, values, and perceptions—and systematically maps the empirical methods, data types, and evaluation metrics used to assess these facets. The findings reveal gaps in facet coverage, highlight underutilized empirical designs, and stress the importance of quality assessment and methodological triangulation. This research offers a facet-wise synthesis of empirical designs and provides actionable guidelines to help researchers and practitioners empirically evaluate Dev-X. Conclusion: Based on SLR findings, this paper presents a Dev-X facet-based synthesis of empirical designs to guide selection of method, data, and metric combinations in future studies. By revealing design gaps (e.g., underexplored facets, overuse of surveys, weak corroboration) and highlighting dominant practices (e.g., survey-based ratings for emotions and values), our facet-wise schema offers an evidence-based guide for context-appropriate Dev-X evaluation strategies. Editor’s note: Open Science material was validated by the Journal of Systems and Software Open Science Board .

We tackle the problem of computing guarantees for the robustness of neural networks against quantisation of their inputs, parameters and activation values. In particular, we pose the problem of bounding the worst-case discrepancy between the original neural network and all possible quantised ones parametrised by a given maximum quantisation diameter $\epsilon > 0$ over a finite dataset. To achieve this, we first reformulate the problem in terms of bilinear optimisation, which can be solved for provable bounds on the robustness guarantee. We then show how a quick scheme based on interval bound propagation can be developed and implemented during training so to allow for the learning of neural networks robust against a continuous family of quantisation techniques. We evaluated our methodology on a variety of architectures on datasets such as MNIST, F-MNIST and CIFAR10. We demonstrate how non-trivial bounds on guaranteed accuracy can be obtained on several architectures and how quantisation robustness can be significantly improved through robust training.

While offering a principled framework for uncertainty quantification in deep learning, the employment of Bayesian Neural Networks (BNNs) is still constrained by their increased computational requirements and the convergence difficulties when training very deep, state-of-the-art architectures. In this work, We reinterpret weight-sharing quantization techniques from a stochastic perspective in the context of training and inference with Bayesian Neural Networks (BNNs). Specifically, we leverage 2D-adaptive Gaussian distributions, Wasserstein distance estimations, and alpha-blending to encode the stochastic behavior of a BNN in a lower-dimensional, soft Gaussian representation. Through extensive empirical investigation, we demonstrate that our approach significantly reduces the computational overhead inherent in Bayesian learning by several orders of magnitude, enabling efficient Bayesian training of large-scale models, such as ResNet-101 and Vision Transformer (VIT). On various computer vision benchmarks—including CIFAR-10, CIFAR-100, and ImageNet1k—our approach compresses model parameters by approximately 50$\times$ and reduces model size by 75% while achieving accuracy and uncertainty estimations comparable to state-of-the-art.

Genetic algorithms (GA) allow us to optimise graphs according to multiple objectives while considering many different constraints. These population-based algorithms assess the fitness of a high number of genomes. In the case of optimising road networks, a high number of fitness assessments leads to high computational costs of traffic simulations. In this work, we explore the application of learning graph configuration spaces to make efficient use of these traffic simulations by using learning model predictions for the majority of fitness assessments. In a controlled experiment, we compare the quality of GA optimisations with and without learning model predictions on the same simulation budget. Our results indicate that although we lose accuracy in the fitness assessments with predictions, the GA reliably finds road networks with better traffic flow and lower overall road length while using the same number of traffic simulations. We show that learning models can support GAs to make efficient use of the simulation budget and thus improve the optimisation. Future work is necessary to confirm these results for larger road networks.

The study presented in this paper aims to provide evidence for the hypothesis that software product line research has been changing and that the works in industry and academia have diverged over time. We analysed a subset (140) of all (593) papers published at the Software Product Line Conference (SPLC) until 2017. The subset was randomly selected to cover all years as well as types of papers. We assessed the research type of the papers (academic or industry), the kind of evaluation (application example, empirical, etc.), and the application domain. Also, we assessed which product line life-cycle phases, development practices, and topics the papers address. We present an analysis of the topics covered by academic vs. industry research and discuss the evolution of these topics and their relation over the years. We also discuss implications for researchers and practitioners. We conclude that even though several topics have received more attention than others, academic and industry research on software product lines are actually rather in line with each other.

This paper summarizes the motivation, objectives, and format of the 5th International Workshop on Product LinE Approaches in Software Engineering (PLEASE15). The main goal of the PLEASE workshop series is to encourage and promote the adoption of Software Product Line Engineering. This year's edition focuses on the link between software product line engineering (SPLE) and new challenges posed by emerging societal trends. Towards this end, we invited reports on (1) opportunities posed by societal challenges for SPLE research and practice and (2) concrete solutions exemplifying application of SPLE techniques to societal challenges.

Expert software engineers working on large systems often need to perform feature location when moving to work in unfamiliar areas. We hypothesise that leveraging the system-specific knowledge of these may help to improve semi-automated feature location techniques. In order to assess and understand how software nomads perform manual feature location searches, two expert professional software engineers were observed in-vivo following a think-aloud protocol while performing manual feature location on a large-scale heterogeneous system. The nomads' search actions were found to be around twice as effective as those reported in previous studies. This cannot be explained by sophisticated use of tools or complex queries. We conclude that system rules and conventions are frequently used by experts when constructing feature location search terms.

Background. In order to understand research on a particular computing topic, practitioners and researchers often need to obtain an overview of its research methods. Current research methods coding schemes either capture insufficient details to support a full critical assessment, or are specialised to a particular research type. Aim. This paper defines and demonstrates RSML, a Research Schema Modelling Language that captures a high level of detail and is applicable to most types of computing research. Method. RSML was designed using concepts from the research methods literature, then refined inductively. An RSML editor was created to assist coders and help reduce coding errors. To demonstrate the feasibility of modelling research with RSML, and to exemplify the summary information that can be derived from a database of RSML encodings, a trial review of 24 articles from one journal was conducted. Results. The review illustrates quantitatively the journal's focus on artifact construction and empiricism. It also reveals that observations are rarely used to inform artifact construction, and purely empirical studies are scarce. Conclusion. RSML can be used to model sophisticated, multifaceted research spanning a wide range of software engineering topics, yielding insights that are not easily captured by current coding schemes.

Feature-models (fms) are a widely used approach to specify the commonalities and variability in variable systems and software product lines. Various works have addressed edits to fms for fm evolution and tool support to ensure consistency of fms. An important extension to fms are feature cardinalities and related constraints, as extensively used e.g., when modeling variability of cloud computing environments. Since cardinality-based fms pose additional complexity, additional support for evolution and consistency checking with respect to feature cardinalities would be desirable, but has not been addressed yet. In this paper, we discuss common cardinality-based fm edits and resulting inconsistencies based on experiences with fms in cloud domain. We introduce tool-support for automated inconsistency detection and explanation based on an off-the-shelf solver. We demonstrate the feasibility of the approach by an empirical evaluation showing the performance of the tool.

This paper summarizes PLEASE 2013, the Fourth International Workshop on Product LinE Approaches in Software Engineering. The main goal of PLEASE is to encourage and promote the adoption of Software Product Line Engineering. To this end, we aim at bringing together researchers and industrial practitioners involved in developing families of related products in order to (1) facilitate a dialogue between these two groups and (2) initiate and foster long-term collaborations.

One of the greatest barriers on the way to the efficient creation, handling, and evolution of product lines is the complexity and scale of the underlying artifacts. In this context, the MAPLE/SCALE workshop focuses on the investigation of scalability issues and the application of model-driven concepts and techniques in software product line engineering (SPLE). The workshop explores how to handle product lines of realistic complexity and how to facilitate systematic and efficient product derivation.

The MAPLE workshop focuses on the application of concepts and techniques from Model-driven Software Engineering (MDSE) in Software Product Line Engineering (SPLE). A particular focus is on techniques that allow to derive products from a product line more efficiently in order to maximize the return on investment and realize the expected benefits of SPLE techniques. The workshop covers both automated and interactive model-based techniques. As submissions, we particularly encourage research papers based on industrial experience and empirical studies as well as contributions, which identify and structure open challenges and research questions.

Software Product Lines (SPL) enable efficient derivation of products. SPL concepts have been applied successfully in many domains including interactive applications. However, the user interface (UI) part of applications has barely been addressed yet. While standard SPL concepts allow derivation of functionally correct UIs, there are additional non-functional requirements, like usability, which have to be considered. This paper presents a case study investigating UI variability found in variants of the commercial web-based information system HIS-GX/QIS. We analyze which aspects of a UI vary and to which degree. The results show that just tweaking the final UI (e.g., using stylesheets) is not sufficient but there is a need for more customization which must be supported by, e.g., UI-specific models.

We consider the problem of supporting effective code reuse as part of Software Product Line Engineering. Our approach is based on code forking -- a practice commonly used in industry where new products are created by cloning the existing ones. We propose to maintain meta-information allowing organization to reason about the developed product line in terms of features rather than incremental code changes made in different forks and to detect inconsistencies in implementations of these features. In addition, we propose to detect and maintain semantic, implementation-level require relationships between features, supporting the developers when they copy features from different branches or delete features in their own branch, thus facilitating reuse of features between products. Our approach aims at mitigating the disadvantages of the forking mechanism while leveraging its advantages. We illustrate the approach on an example, and discuss its possible implementation and integration with Software Configuration Management systems.

PLEASE workshop series focuses on exploring the present and the future of Software Product Line Engineering techniques. The main goal of PLEASE 2011 is to bring together industrial practitioner and software product line researchers in order to couple real-life industrial problems with concrete solutions developed by the community.

Many of the benefits expected from software product lines (SPL) [1-2] are based on the assumption that the additional investment required for domain engineering, pays off during application engineering when products are derived from the product line [3]. However, to fully exploit this we need to optimize application engineering processes and handle the reusable artifacts of an SPL in a systematic and efficient manner. In this context, the joint MAPLE/SCALE workshop focuses on two closely related aspects: how model-driven approaches can help to achieve systematic and efficient derivation of products and how scalability challenges can be addressed that arise from the application of SPL techniques to SPLs of realistic size and complexity. The workshop aims to explore and explicate the current status and ongoing work in model-driven approaches and/or scalability of SPLs and the transfer of knowledge between different disciplines and application domains.

In industrial settings, products are rarely developed by one organization alone. Software vendors and suppliers typically maintain their own product lines, which can contribute to a larger (multi) product line. The teams involved often use different approaches and tools to manage the variability of their systems. It is unrealistic to assume that all participating units can use a standardized and prescribed variability modeling technique. The configuration of products based on several models in different notations and with different semantics is not well supported by existing approaches. In this paper we present an integrative approach that provides a unified perspective to users configuring products in multi product line environments, regardless of the different modeling methods and tools used internally. We also present a technical infrastructure and a prototypic implementation based on Web Services. We show the feasibility of the approach and its implementation by using it with two different variability modeling approaches (one feature-based and one decision-oriented approach) on an example derived from industrial experience.

PLEASE is a new workshop series that focuses on exploring the present and the future of Software Product Line Engineering (SPLE) techniques. The goal of the workshop is to bring together researchers and practitioners with special interest in SPLE in order to discuss ongoing research and new ideas for advancing the field. The workshop's main theme, Beyond Product Lines, focuses on the adaptation of SPLE to dynamic settings in which neither the goal nor the organizational structure is stable.

The initial vision associated with multimedia is to achieve user interfaces (UI) which are highly interactive, intelligent, usable, and entertaining. Although early multimedia applications did not always fulfill these expectations, today's applications come much closer towards this vision. Examples are Rich Internet Applications like Google Maps, Edutainment Systems, or new UIs in the context of Ubiquitous Computing. The key difference is the increased dynamics of the UI (media objects are created/modified depending to the current context) and the tighter integration of UI elements with the application logic. Although this dynamism allows new forms of applications, it is problematic when engineering such applications: Existing multimedia modeling approaches focus mainly on the spatio-temporal integration of media objects. However, they hardly allow modeling the integration with the application logic or generic media objects. Hence, many of these concepts cannot be handled. To address these challenges, this paper first illustrates the problem with examples from common applications. Subsequently, we propose a solution how to integrate these concepts into a consistent modeling language. The discussed issues can be generalized from the presented language and integrated into other approaches, e.g., modeling languages for Rich Internet Applications.

Software Product Line engineering has emerged as a viable and important software development paradigm in the automotive industry. It allows companies to realise significant improvements in time-to-market, cost, productivity, and system quality. One major difficulty with software product line engineering is related to the fact that a product line of industrial size can easily incorporate thousands of variation points. This scale of variability can become extremely complex to manage resulting in a product configuration process that bears significant costs. This paper introduces a meta-model and research tool that employs visualisation and interaction techniques to improve product configuration in high-variability product lines. The meta-model and techniques utilised are illustrated using an automotive restraint system example.

Software product lines of industrial size can easily incorporate thousands of variation points. This scale of variability can become extremely complex to manage resulting in a product development process that bears significant costs. One technique that can be applied beneficially in this context is visualisation. Visualisation is widely used in software engineering and has proven useful to amplify human cognition in data intensive applications. Adopting this technique in software product line engineering can help stakeholders in supporting essential work tasks and in enhancing their understanding of large and complex product lines. This paper introduces a meta-model and accompanied research tool that employs visualisation techniques to support fundamental product line development tasks. The meta-model specifies major entities such as decisions, features, and components and relationships among them. We discuss which tasks can be supported based on the meta-model and show examples of how these tasks can be further enhanced by utilising interactive visualisation techniques implemented in our tool.

Product Derivation is one of the central activities in Software Product Lines (SPL). One of the main challenges of the process of product derivation is dealing with complexity, which is caused by the large number of artifacts and dependencies between them. Another major challenge is maximizing development efficiency and reducing time-to-market, while at the same time producing high quality products. One approach to overcome these challenges is to automate the derivation process. To this end, this paper focuses on one particular activity of the derivation process; the derivation of the product-specific architecture and describes how this activity can be automated using a model-driven approach. The approach derives the product-specific architecture by selectively copying elements from the product-line architecture. The decision, which elements are included in the derived architecture, is based on a product-specific feature configuration. We present a prototype that implements the derivation as a model transformation described in the Atlas Transformation Language (ATL). We conclude with a short overview of related work and directions for future research