A Runtime Framework for Machine-Augmented Software Design Using Unsupervised Self-Learning

Abstract

Modern computer software comprises tens of millions of lines of code and is deployed in highly dynamic environments such as data-centres, with constantly fluctuating user populations and popular content patterns. Together this complexity and dynamism make computer software very difficult to develop and maintain. The autonomic computing community has grown to address some of these challenges, developing automation in areas such as self-optimisation and self-healing. However, work to date either (i) focuses on a specific problem in isolation, neglecting the broader complexity of software construction, or (ii) considers the design process but is human-centric, relying on expertly-crafted models. In this paper we examine software development as a process, infusing this process with a level of autonomy that seeks to make software an active part of its own development team. We present an overview of our framework and we demonstrate the accuracy of our framework in autonomously finding the most suitable software design at runtime according to specific operating conditions.