Choosing a cloud-based machine learning platform

Published on Oct 11, 2022

These are 12 capabilities every cloud machine learning platform should have to support the complete machine learning lifecycle.

For machine learning and deep learning to be effective, you need copious amounts of data, a way to clean it and perform feature engineering, and a way to train the models quickly. Next, you need a way to deploy your models, monitor them over time, and retrain them.

The same can be done on-premises with compute resources and accelerators such as GPUs, but if your resources are adequate, you may find that they are idle much of the time. However, in some cases it can be more cost-effective to run the entire pipeline in the cloud, using large amounts of compute resources and accelerators as needed.

Major cloud providers – and many minor clouds too – have built out their machine learning platforms to support the complete machine learning lifecycle, from planning to maintaining models. Which cloud is right for you? These are 12 capabilities every end-to-end machine learning platform should have, with notes on which clouds provide them.

When you have large amounts of data to build precise models, you don’t want to ship it halfway around the world. Data transmission latency is ultimately limited by the speed of light, even on an infinite bandwidth network. Distance means latency.

For very large data sets, it is ideal to build the model where the data already resides, so no mass data transfer is required. Databases support it.

Data should be on the same high-speed network as the model-building software, which typically means within the same data center. When you have terabytes (TB) or more, moving data between data centers within a cloud availability zone can introduce a significant delay. Incremental updates can mitigate this.

A constrained bandwidth and high latency path would be the worst case for moving big data long distances. Australia’s trans-Pacific cables are particularly egregious.

Cloud providers have addressed this issue in multiple ways. Add machine learning and deep learning to their database services. Using SQL commands against Amazon Redshift, a managed, petabyte-scale data warehouse service, Amazon Redshift ML allows SQL users to create, train, and deploy machine learning models. Google Cloud’s managed, petabyte-scale data warehouse, BigQuery ML, lets you create and run machine learning models.

IBM Db2 Warehouse on Cloud supports in-database SQL analytics, machine learning, and R and Python. SQL Server Machine Learning Services supports R, Python, Java, the PREDICT T-SQL command, the rx_Predict stored procedure, and Spark MLlib in SQL Server Big Data Clusters. Oracle Cloud Infrastructure (OCI) Data Science is a managed and serverless platform to build, train, and manage machine learning models using Oracle Cloud Infrastructure.

Another way cloud providers have addressed this issue is to bring their cloud services into customer data centers as well as to satellite points of presence (often in large metropolitan areas) that are closer to customers. Microsoft Azure calls these Azure Stack Edge nodes and Azure Arc; Google Cloud Platform calls them network edge locations, Google Distributed Cloud Virtual, and Anthos on-premise.