At the recent Ignite conference, Microsoft released several updates related to its Artificial Intelligence (AI) services and tools. These updates include the release of the Azure ML Experimentation service, Azure ML Model Management service, Azure ML Workbench and the general availability of Microsoft Cognitive Services.
Microsoft has seen considerable adoption of their Machine Learning platform, but have identified some opportunities for the service. Matt Winkler, group program manager at Microsoft, explains:
We have hundreds of thousands of models that have been deployed and serve billions of requests. But, we have been doing this for a few years now and are starting to see some recurring patterns. Customers have told us they love the convenience. But have also told us they need greater control over compute and data with more options for model deployment. Customers also have very diverse needs for frameworks, but want the same capabilities for the management and deployment of those models.
New capabilities are expected to address these customer identified gaps by giving more control to customers over how their models are developed and deployed.
Image Source: https://myignite.microsoft.com/videos/55281
Azure ML Experimentation Service
The Azure ML Experimentation Service uses Git-based checkpointing and version control to manage project dependencies and training jobs, whether executed locally or in a scaled-up or scaled-out fashion. In addition, data scientists have the ability to choose their own framework, including TensorFlow, Microsoft CNTK or SparkML. They also can choose their preferred tool including Microsoft Code, Visual Studio, Jupyter or PyCharm. The service also captures service-side run metrics, output logs and models.
Image Source: https://myignite.microsoft.com/videos/55281
Azure ML Model Management Service
The Model Management service provides customers with the control and flexibility of where and how they want to deploy their models. By utilizing Docker, customers have a lot of portability of deploying their models on-premises or in the cloud.
The deployment and management of models is achieved through HTTP endpoints. Application Insights can be used for monitoring and insight into the performance of the models. First class support for SparkML, Python, Cognitive Toolkit, TF, R is included and extensible to support other tools like Caffe and MXnet.
Image Source: https://myignite.microsoft.com/videos/55281
Azure ML Workbench
The Azure ML Workbench is a Windows and Mac OSX tool for AI development. It includes a full environment setup for Python and Jupyter and includes embedded notebooks. It also provides an extensive run history and an experiment comparison experience. Microsoft has also invested in data wrangling tools that simplify the experience of importing data into your data science experiments. This data wrangling capability includes sampling and understanding the data, then performs transformations on top of it. This is enabled through PROSE which is a data preparation by example technology.
Microsoft Cognitive Services
Microsoft also announced the update of their Cognitive Services platform, including the general availability of the Text Analytics Service. The Text Analytics service allows developers to perform sentiment analysis, detect key phrases, topics and language from text.
