Steve Counsell’s research while affiliated with Brunel University London and other places

... Besides, digital transformation enhances internal controls by automating processes that, in the analogue era, used to be prone to human error or manipulation (Suri, 2022). Blockchain technology, for instance, guarantees data integrity and complete transparency, making it more difficult for fraudsters to tamper with financial records without detection (Dashkevich et al., 2024). According to Tariq et al. (2022), RPA and AI have introduced a number of automated processes that close all the entry points created due to manual intervention. ...

Reference:

Exploring the Role of Digital Transformation in Mitigating Accounting Fraud: A Cybersecurity Perspective

... Early work by Wilander and Kamkar [21] showed that five tools missed most C function vulnerabilities and produced many false positives, a trend later echoed by Emanuelsson and Nilsson [22]. Johns and Jodeit [23] demonstrated synthetic benchmarks to distinguish genuine alerts from false alarms, while Bennett [24] reported detection rates of 11.2%-26.5% for standard SAST tools, improved to 44.7% by augmenting them with enhanced Semgrep rules. ...

Reference:

CASTLE: Benchmarking Dataset for Static Code Analyzers and LLMs towards CWE Detection

... Software engineers are found to be well-educated and well-paid [31]. Therefore, like related studies comparing female and male software engineers [34], we also question the necessity for gender comparisons in the field of software engineering. ...

Reference:

Interest in Working Remotely: Is Gender a Factor?

... As described before, while the technical complexity and performance of automated debugging techniques has been increasing (Jiang et al. 2023b), including the use of LLMs for APR (Jiang et al. 2023a;, empirical work on explaining results for developer consumption has been difficult to identify. In addition to Monperrus' living review on APR having only one paper mentioning explanations (Monperrus 2020), Winter et al. (2022) find 17 human studies evaluating APR, of which none involved explanations directly from an APR tool; Kochhar et al. (2016) survey fault localization techniques at the time, and find two techniques that could provide explanations of their results (Sun and Khoo 2013;Mariani et al. 2011); unfortunately, both papers did not have human studies. ...

Reference:

Explainable automated debugging via large language model-driven scientific debugging

... The ultimate goal of Automated Program Repair (APR) pipelines [1] is identify a faulty code fragment, generate a patch, validate it, and ultimately propose it to a human developer, who will either discard or accept it. In this respect, two dimensions are important: speed of patch generation [2] and a (small) number of quality patches surviving the process [3]. Williams et al. [4] showed that by improving the way, time, and context in which APR patches are suggested significantly increased their adoption rate at Bloomberg. ...

Reference:

Using ML filters to help automated vulnerability repairs: when it helps and when it doesn't

... Unlike code-level technical debt, which often manifests in clear, tangible issues such as poor naming or code duplication [11], ATD operates at the high-level design and structure of the software [12]. It is inherently more abstract, often embedded in architectural decisions and dependency relationships. ...

Reference:

Tracing the Lifecycle of Architecture Technical Debt in Software Systems: A Dependency Approach

... The results demonstrated that using properties alongside unit tests can reduce overfitting and improve repair effectiveness. Winter et al. [15] emphasized the importance of incorporating software developers' perspectives and needs into APR technologies' research and development to bridge the gap between technical advancements and practical usability. ...

Reference:

DALO-APR: LLM-based automatic program repair with data augmentation and loss function optimization

... Learning-based end-to-end repair techniques predict patches for buggy programs by learning features of the faulty code sections as well as their correct (developer-written) fixes Chen et al. (2022). Among APR approaches, heuristicbased APR has seen the earliest and thus far most industrial uptake, including being first applied in the context of a management system for a medical application Haraldsson et al. (2017), or more recently being used for automated end-toend repair at scale at Meta Marginean et al. (2019) or targeting frequently occurring bugs at Bloomberg Kirbas et al. (2020). ...

Reference:

Reinforcement learning for mutation operator selection in automated program repair

... A code smell is an indicator that characterizes bad practices, design problems, or potential bugs in code [1], which may lead to code violations or vulnerability, and thus increase the possibility of project failure. Studying code smells is important, but little research compares their distribution and evolution between different programming languages [2]. ...

Reference:

Towards Understanding the Distribution and Evolution of Code Smells: An Empirical Study on Java and Python Projects (S)

... As far as possible, the workshop kept the traditional (physical) format, starting with an invited keynote given by Prof. Mark Harman, who described the use of SBSE [1] and Genetic Improvement [ [11], within Facebook and future plans, including social testing [12] and Facebook calls for research proposals. This was followed by formal (albeit electronic only) presentations of papers, which are to be published in the ACM digital library [12,13,14,15,16,17]. (The keynote and all of the presentations were recorded and are available via YouTube https://youtu.be/GsNKCifm44A). ...

Reference:

Genetic Improvement @ ICSE 2020