The main challenges are discussed together with the lessons learned from past and ongoing research along the development cycle of machine learning systems. This will be done by taking into account intrinsic conditions of nowadays deep learning models, data and software quality issues and human-centered artificial intelligence (AI) postulates, including confidentiality and ethical aspects. The analysis outlines a fundamental theory-practice gap which superimposes the challenges of AI system engineering at the level of data quality assurance, model building, software engineering and deployment. The aim of this paper is to pinpoint research topics to explore approaches to address these challenges.

In the last decade, industry’s demand for deep learning (DL) has increased due to its high performance in complex scenarios. Due to the DL method’s complexity, experts and non-experts rely on blackbox software packages such as Tensorflow and Pytorch. The frameworks are constantly improving, and new versions are released frequently. As a natural process in software development, the released versions contain improvements/changes in the methods and their implementation. Moreover, versions may be bug-polluted, leading to the model performance decreasing or stopping the model from working. The aforementioned changes in implementation can lead to variance in obtained results. This work investigates the effect of implementation changes in different major releases of these frameworks on the model performance. We perform our study using a variety of standard datasets. Our study shows that users should consider that changing the framework version can affect the model performance. Moreover, they should consider the possibility of a bug-polluted version before starting to debug source code that had an excellent performance before a version change. This also shows the importance of using virtual environments, such as Docker, when delivering a software product to clients.

Software modernization is a common activity in software engineering, since technologies advance, requirements change, and business models evolve. Differently from conventional software evolution (e.g., adding new features, enhancing performance, or adapting to new requirements), software modernization involves re-engineering entire legacy systems (e.g., changing the technology stack, migrating to a new architecture style, or programming paradigms). Given the pervasive nature of software today, modernizing legacy systems is paramount to provide customers with competitive and innovative products and services, while keeping companies profitable. Despite the prevalent discussion of software modernization in gray literature, and the many papers in the literature, there is no work presenting a “big picture” of contemporary software modernization, describing challenges, and providing a well-defined research agenda. The goal of this work is to describe the state of the art in software modernization in the past 10 years. We collect the state of the art by performing a rapid review (searching five digital libraries), identifying potential 3,460 studies, leading to a final set of 126. We analyzed these studies to understand which strategies are employed, the driving forces that lead organizations to modernize their systems, and the challenges that need to be addressed. The results show that studies in the last 10 years have explored eight strategies for modernizing legacy systems, namely cloudification, architecture redesign, moving to a new programming language, targeting reuse optimization, software modernization for new hardware integration, practices to leverage automation, database modernization, and digital transformation. Modernization is triggered by 14 driving forces, with the most common ones being reducing operational costs, improving performance and scalability, and reducing complexity. In addition, based on the analysis of existing literature, we present a detailed discussion of research opportunities in this field. The main challenges are providing tooling support, followed by defining a modernization process and considering better evaluation metrics. The main contribution of our work is to equip practitioners and researchers with knowledge of the current state of contemporary software modernization so that they are aware of practices and challenges to be addressed when deciding to modernize legacy systems.