LinuxCon ContainerCon CloudOpen China

Running large-scale machine learning is frustrating. It requires massive data and computational power. Its model is often built with frameworks like Tensorflow, Caffe, SparkML. Each fairly requires big efforts to manage libraries and dependencies. It commonly takes a long procedure from preparing data, training model to online prediction. In Aliyun, Kai's team builds elastic machine learning platform based on Docker technology to resolve above difficulties. It provides services to support environment portability and repeatability, CPU/GPU isolation, job scheduling, monitoring, load balancing and auto scaling.
In this topic, Kai introduces how they facilitate end-to-end lifecycle of large-scale machine learning workload with GPU acceleration through container management and orchestration. Demo will show how people can start to train his first neural network model in minutes.