Microsoft's cloud machine learning service ups the stakes against AWS and Google in a race to provide machine learning to the masses through ease of use and support for popular frameworks.
Microsoft's Azure Machine Learning service is now generally available for developers and data scientists who look for more efficient ways to build machine learning models. Microsoft's disclosure comes just days after AWS released enhancements to its SageMaker machine learning platform at the AWS re:Invent 2018 conference, a platform that helps developers and data scientists build, train and deploy machine learning models.
"Microsoft is keeping pace with Amazon and Google with [its] enhanced support for AI and machine learning capabilities within [its] Visual Studio Code development environment and [its] cloud-based ML platform," said Ronald Schmelzer, an analyst at Cognilytica, based in Ellicott City, Md.
Tooling and frameworks support
The Azure Machine Learning service speeds up the process of identifying useful algorithms and machine learning pipelines, which automates model selection and tuning. This can cut development time from days to hours, said Bharat Sandhu, director of product marketing, big data and analytics at Microsoft.
It also provides DevOps capabilities, via integrated CI/CD tooling, to enable experiment tracking and manage machine learning models deployed in the cloud and on the edge, said Venky Veeraraghavan, group program manager for Microsoft Azure, in a blog post.
The Azure Machine Learning service supports popular open source frameworks, including PyTorch, TensorFlow and scikit-learn, so developers and data scientists can use familiar tools. Developers can use Visual Studio Code, Visual Studio, PyCharm, Azure Databricks notebooks or Jupyter notebooks to build apps that use the service.
Cloud providers cast nets for ML coders
Microsoft, AWS and Google have increased their outreach to users that want to use machine learning in a low-risk, high-return manner but not require in-house proficiency or paid experts.
Holger Muelleranalyst, Constellation Research
"When a vendor needs to catch up to the leader -- or leaders -- as Microsoft has to in AI, then it needs to do this with two strategies: ease of use and breadth of offering," said Holger Mueller, an analyst at Constellation Research, based in San Francisco. "Microsoft is doing both -- making it easier for AI specialists and data scientists to bring the data together, create, validate and deploy AI models on the one dimension. On the other dimension, it offers the wide range of languages and models for use cases beyond the very popular TensorFlow."
The Azure Machine Learning service automates feature extraction, algorithm selection and hyper-parameter tuning to optimize learning algorithms, Schmelzer said.
"This is great for less experienced data science and machine learning engineers who might have a handle on data, but not as much on the specifics of different ML algorithms or their configurations," he said.
Azure Machine Learning also incorporates notebooks into the Visual Studio Code environment, to allow side-by-side code and data science activities, which help to make machine learning a better citizen in the Microsoft code environment, he said.
By comparison, AWS' machine learning suite targets casual users and provides better support for higher-level ML activities, such as forecasting and personalization, as well as a broader range of capabilities around natural language, data wrangling and more, Schmelzer noted.
"Amazon continues to widen its suite and appeal to broader user types, whereas Microsoft's ML approach still seems focused on developers and data scientists rather than casual business users or low code developers," he said.
All eyes will now be on adoption of PyTorch and the Open Neural Network Exchange (ONNX), an open format for deep learning and machine learning models developed by Microsoft, Facebook and AWS, Mueller said. Microsoft has released the source code for the ONNX Runtime, an inference engine for machine learning models in the ONNX format that runs on Windows, Linux and Mac. Microsoft teams use the ONNX Runtime to improve the scoring latency for machine learning models in Bing Search, Bing Ads and Office productivity services.
"The next months will tell if enterprises will use Microsoft Azure more for their AI/ML projects," Mueller said.