Azure Machine Learning—ML for all skill levels

Enterprises today are adopting artificial intelligence (AI) at a rapid pace to stay ahead of their competition, deliver innovation, improve customer experiences, and grow revenue. AI and machine learning applications are ushering in a new era of transformation across industries from skill sets to scale, efficiency, operations, and governance.

Microsoft Azure Machine Learning provides enterprise-grade capabilities to accelerate the machine learning lifecycle and empowers developers and data scientists of all skill levels to build, train, deploy, and manage models responsibly and at scale. At Microsoft Ignite, we’re announcing a number of major advances to Azure Machine Learning across the following areas:

◉ New studio web experience that boosts machine learning productivity for developers and data scientists of all skill levels, with flexible authoring options from no-code drag-and-drop and automated machine learning, to code-first development.

◉ New industry-leading Machine Learning Operations (MLOps) capabilities to manage the machine learning lifecycle, enabling data science and IT teams to deliver innovation faster.

◉ New open and interoperable capabilities that provide choice and flexibility with support for R, Azure Synapse Analytics, Azure Open Datasets, ONNX, and other popular frameworks, languages, and tools.

◉ New security and governance features including role-based access control (RBAC), Azure Virtual Network (VNet), capacity management, and state-of-the-art responsible AI interpretability and fairness capabilities.

Let’s dive into these announcements in detail to see how Azure Machine Learning is helping individuals, teams, and organizations meet and exceed business goals.

Access machine learning for all skill levels and boost productivity

The new studio web experience (currently in preview) enables data scientists and data engineers of all skill levels to complete end-to-end machine learning tasks, including data preparation, model training, deployment, and management in a seamless manner. Choose from three different authoring options based on your skill and preference—no-code drag-and-drop designer, automated machine learning, or a code-first notebooks experience. Access Azure Machine Learning assets (including datasets and models) and rich capabilities (including data drift, monitoring, labeling and more) all from a single location.

Studio web experience

Designer (currently in preview) provides drag-and-drop workflows to simplify the process of building, testing, and deploying machine learning models using a visual experience. Customers currently using the classic version of Azure Machine Learning Studio are encouraged to try Designer so they can benefit from the scale and security of Azure Machine Learning.

Automated machine learning user interface (currently in preview) helps data scientists build models without writing a single line of code. Automate the time-intensive tasks of feature engineering, algorithm selection, and hyperparameter sweeping, then operationalize your model with a few clicks of a button.

Notebooks (currently in preview) are a fully managed solution for developers and data scientists to easily get started with machine learning, with pre-configured custom environments that eliminate setup time, while providing management and enterprise readiness capabilities for IT administrators.

New data labeling (currently in preview). High quality labeled data is vital to creating high accuracy models for supervised learning. Teams can now manage data labeling projects seamlessly from within the studio web experience to get labels against data, speeding up the time-intensive process of manual labeling. Labeling tasks supported include object detection, multi-class image classification, and multi-label image classification.

Operationalize at scale with industry-leading MLOps

Azure Machine Learning features built-in MLOps capabilities for enterprise-grade machine learning lifecycle management, that enables data science and IT teams to collaborate and increase the pace of model development and deployment.

New updates to build reproducible models and achieve machine learning governance and control

Datasets help data scientists and machine learning engineers easily access data from a number of Azure storage services, apply datasets rapidly, reuse them efficiently across tasks, and track data lineage automatically. Rich dataset and model registries help track assets and information to effectively operationalize models and simplify workflows from training to inferencing. Version control helps track and manage assets providing enhanced traceability and supporting the creation of reproducible pipelines for consistent model delivery. Audit trail capabilities ensure asset integrity and provide control logs to help meet regulatory requirements.

New updates to easily deploy models and efficiently manage the machine learning lifecycle

Batch inference helps increase productivity and decrease cost by generating predictions on terabytes of structured or unstructured data. Controlled roll-out enables the deployment of different model versions under a common scoring endpoint in order to implement a sophisticated deployment pipeline and release models with confidence. Data drift monitoring helps maintain model accuracy by detecting model performance issues from changes to model input data over time. Drift analysis includes magnitude of drift, contribution by feature, and other insights so that appropriate action can be taken, including retraining the model.

Data drift monitoring

Innovate using open and interoperable capabilities

With Azure Machine Learning, developers and data scientists can access built-in support for open source tools and frameworks like PyTorch, TensorFlow, and scikit-learn, or the open and interoperable ONNX format. We now support Open Neural Network Exchange (ONNX), the open standard for representing machine learning. With the new v1.0 release, ONNX Runtime offers stable Python APIs that can be used in Azure Machine Learning on both CPU and GPU.

New R-based capabilities enable data scientists to run R jobs on Azure Machine Learning and then manage and deploy R models as web services. Data scientists can choose their development environment of choice—one-click access to the browser integrated development (IDE) of RStudio Server (open source edition) or Jupyter with R.

Azure Synapse Analytics is now deeply integrated with Azure Machine Learning to greatly expand the discovery of insights from all your data and apply machine learning models to your intelligent apps.

Azure Open Datasets are now generally available and provide curated datasets, hosted on Azure, and easily accessible from Azure Machine Learning workspaces to accelerate model training. Over 25 datasets are now available, including socio-economic data, satellite imagery, and more. New datasets are continuously being added, and you can nominate additional datasets to Azure.

Build on a secure foundation

Security and enterprise readiness updates

Workspace capacity management (currently in preview) helps administrators review compute usage across workspaces and clusters within a subscription for efficient resource distribution. Capacity limits can be set to reallocate resources for capacity management and governance. Role Based Access Control, or RBAC, (in preview) helps define custom roles for granular access control and supports advanced security scenarios. Virtual network, or VNet, (in preview) provides a security boundary to isolate compute resources used to train and deploy models when running experiments through inferencing.

Fairness: In addition to model interpretability in Azure Machine Learning, which supports transparency and model understanding, data scientists and developers can now leverage Fairlearn, the new open source fairness assessment and mitigation tool. This tool assists organizations with uncovering insights about fairness in their model predictions through an intuitive and configurable set of visualizations.

Fairness feature insights