Powering your application with deep learning is no walk in the park, but is certainly attainable with some tricks & good practice. Serving a deep learning model on a production system demands the model to be stable, reproducible, capable of isolation & to behave as a stand-alone package. One possible solution to this is a containerized microservice.|
Ideally, serving deep learning microservices should be quick & efficient, without having to dive deep into the underlying algorithms & their implementation. Too good to be true? Not anymore! Together, we will demystify the process of developing, training, & deploying deep learning models as a web microservice using Model Asset Exchange, an open source framework developed in Python at the IBM Center for Open Source Data & AI Technologies (CODAIT).
We will kick off with an overview of how deep learning models are best published as Docker Images on DockerHub, & are best prepared for deployment in local or cloud environments using Kubernetes or Docker. We highlight the following benefits of such an approach: Standardized REST API implementation & application-friendly output format (JSON) Abstracting out the complex pre & post-processing portions of the model inputs & outputs.
We will walk you through some super cool applications such as automatic image cropping, age estimation from videos/webcam & Veremin - a video theremin. All these applications & the framework itself are open source & we conclude by inviting contributions & opening the gates for you to be a part of this amazing initiative!