Exploitation Pattern for Machine Learning Systems

Abstract

The rapid increase in the development of machine learning systems in companies today presents new challenges that have not been faced in the development of traditional IT systems. Machine learning systems for human operations have a different set of challenges in comparison with Porter's connected smart products. The methodology for planning and developing such products has not yet been established, and it is mainly based on experience and intuition. In this paper, we define “exploitation” as the process of enhancing machine learning systems and summarize the issues in exploitation pattern based on several case studies. In the latter half of the paper, we propose a development process that minimizes machine learning problems on the exploitation of the systems. This is done by defining multiple uncertainties and proposing multiple development methods that are required for those uncertainties.