In this era, the agile mindset has innovated the traditional software engineering (SE) process through the integration of DevOps flow engines, scrum iterations, and automation of continuous integration (CI) and continuous deployment (CD) cycles. However, the CI/CD integration requires manual code-revisions and refactoring at large scales. Recently, machine-learning (ML) is employed in SE that allows legacy codes to be highly dynamic, less coupling in related modules, automatic code versioning, and refactoring, with less coupling among related modules. However, over time, ML models tend to become bulky, with increasing monotonic losses during model training. To address this, SE techniques like code revisions are employed over ML codes to allow low-order training losses, that enables seamless and precise workflow structures. Motivated from the aforementioned discussions, the paper presents a systematic review of the close interplay of SE and ML and possible interactions in different applications. Suitable research questions and case studies are presented for possible adoption scenarios that depict the close ML-SE interplay share with each other, with the concluding remarks. The paper forms useful insights for ML engineers, data science practitioners, and SE quality estimators towards the building of efficient and scalable software solutions.
Paper
Full text
Interplay of Machine Learning and Software Engineering for Quality Estimations
Semantic Scholar · Computer Science · 2020
Abstract
In this era, the agile mindset has innovated the traditional software engineering (SE) process through the integration of DevOps flow engines, scrum iterations, and automation of continuous integration (CI) and continuous deployment (CD) cycles. However, the CI/CD integration requires manual code-revisions and refactoring at large scales. Recently, machine-learning (ML) is employed in SE that allows legacy codes to be highly dynamic, less coupling in related modules, automatic code versioning, and refactoring, with less coupling among related modules. However, over time, ML models tend to become bulky, with increasing monotonic losses during model training. To address this, SE techniques like code revisions are employed over ML codes to allow low-order training losses, that enables seamless and precise workflow structures. Motivated from the aforementioned discussions, the paper presents a systematic review of the close interplay of SE and ML and possible interactions in different applications. Suitable research questions and case studies are presented for possible adoption scenarios that depict the close ML-SE interplay share with each other, with the concluding remarks. The paper forms useful insights for ML engineers, data science practitioners, and SE quality estimators towards the building of efficient and scalable software solutions.