Deep Learning in Java

Deep Learning in Java
Qing Lan

A presentation from ApacheCon @Home 2020


AI is evolving rapidly, and is used widely in a variety of industries. Machine learning (ML) applications ranging from basic text classification to complex applications such as object detection and pose estimation are being developed to be used in enterprise applications. Currently, software engineers using Java have a large barrier to entry when they try to adopt Deep Learning (DL) for their applications. Python being the de-facto programming language for ML adds additional gradient to an already steep learning curve. This tutorial will introduce an open-source, framework-agnostic Java library — Deep Java Library (DJL) for high-performance training and inference in production. DJL supports a variety of Deep Learning engines (including but not limited to Apache MXNet, TensorFlow and PyTorch) and provides a simple and clean Java API that will work the same with each engine. Additionally, DJL offers the DJL Model Zoo - a repository of models that makes it easy to share models across teams. This tutorial will walk software engineers through the core features of DJL and demonstrate how it can be used to simplify experience of serving models. By the end of the session, users will be able to train and deploy DL models from a variety of DL frameworks into production environments and serve user requests using Java. Website:

Qing is a SDE II in the AWS Deep Learning Toolkits team. He is one of the co-authors of DJL (djl.ai) and PPMC member of Apache MXNet. He graduated from Columbia University in 2017 with a MS degree in Computer Engineering and has worked on model training and inference. Qing has presented a workshop about Apache MXNet in ApacheCon 2019(Las Vegas) about using Java for Deep Learning inference.