The Ecosystem of Machine Learning Methods

Abstract

Machine learning (ML) is a rapidly evolving field and plays an important role in today’s data-driven business environment. Many digital innovations in domains as diverse as healthcare, banking, energy, and retail are powered and enabled by ML. Examples include search engines, recommendation systems, pattern recognition, computer vision, and natural language processing. A key element in ML innovation is the advancement of the underlying methods, which specify how machines should algorithmically process, derive patterns, and learn from data for a given decisioning task. The speed at which this is happening is exponential, with researchers leveraging and building upon existing building blocks as well as introducing entirely new methods. Given the speed, scale, and complexity, understanding this complex evolving ML method space can be challenging. What methods are core and peripheral to ML? Which methods span task areas? How are ML methods evolving? In this exploratory research paper, I address these questions by (1) framing the ML method space and (2) visualizing the evolving structure of the ML methods ecosystem. The results reveal several foundational ML building blocks, different coupling levels between ML areas, and variable speeds of evolution. The study also provides insights into how digital innovation evolves at an algorithmic level. I discuss the implications of the findings and describe opportunities for future ML ecosystem-focused research.