Microsoft has announced the general availability of the Azure Machine Learning service. Azure Machine Learning automates machine learning to make it easier to build, train and deploy models. The service is generally available now, with pricing to go into effect February 1, 2019.
This service automates the process of data transformation, model selection, and hyperparameter tuning. These machine learning solutions can then be deployed in the cloud, on premise, or at the edge (in an IoT scenario). The service includes a software development kit for Python. There are development environments for Visual Studio Code, PyCharm, Azure Databrick notebooks, and Jupyter notebooks.
The distributed deep learning feature enables developers to build and train models with massive clusters of graphical processing units, and to access field programmable gate arrays on Azure for image classifications and recognition.
The service also includes automated model selection and tuning of hyperparameters to potentially eliminate some of the more tedious work in data modeling. Automated model and hyperparameter tuning allows models to self adjust. For example, a recommendation service can adjust the parameters for a given user as it learns what the customer selects.
From a privacy perspective this also can be a benefit because the data remains on the local machine or Azure virtual machine governed by Microsoft's privacy policy. Adjusting the models and parameters could also be useful for operations that have hundreds or thousands of pieces of equipment in various locations to tune the predictive models for each piece of equipment.
The research behind automated machine learning has been published.