2018 has been an excellent year for machine learning, with frameworks such as PyTorch, Karas, Tensorflow and CNTK maturing in levels of stability and ease of use, sophisticated and academic concepts have become highly accessible to a wide range of developers. Anyone investing a few hours in technical tutorials can build a learner; yet, scaled production integration remains a pain point. However, 2018 saw significant advances in Cloud-Native and DevOps technologies. Kubernetes is now as ubiquitous as “the cloud” and with projects like Airflow and KubeFlow integrating seamlessly, along with their resolve to ease ML integration at scale, I expect 2019 to be a big year in highly accessible, vendor-neutral, Cloud-Native Machine Learning.
Declarative Configuration of Machine Learning Architectures
2019 looks to be the year for Cloud-Native Machine Learning. Technologies like Airflow and KubeFlow help reduce complexity and production integration by leveraging declarative configuration and orchestration of data processing pipelines in Kubernetes with a specific focus on Machine Learning. Just as technologies such as PyTorch and Tensorflow brought Machine Learning from academic papers to plug and play libraries, solutions like Airflow and KubeFlow are bringing many frameworks like PyTorch and Tensorflow to enterprise architectures.
Machine Learning Functions as a Service
Kubernetes has propelled the trends in Microservices and FaaS (serverless) architectures by offering an ideal environment, not only in deep native support for their core concepts but additionally easing the integration of CI/CD technologies, propelled by a growing culture of Agile and rapid release schedules. As Microservices and FaaS continue to gain popularity, I expect technologies such as Hydrosphere.io and many others to automate and integrate ML pipelines further into existing Kubernetes based architectures, offering a variety of serverless functions for Machine Learning.
2019 Cloud Native Machine Learning: Airflow and KubeFlow set to ease enterprise deployment of AI/ML by Craig Johnston is licensed under a Creative Commons Attribution 4.0 International License.