Microsoft Azure Machine Learning

At Build 2023, Microsoft today announced several new updates for Azure Machine Learning to improve machine learning professionals’ ability to build and manage AI solutions.

The new Prompt flow preview will offer a streamlined experience for prompting, evaluating and tuning large language models. Using this new service, you can create prompt workflows that connect to language models and data sources. It also integrates Azure AI Content Safety to help users detect and remove harmful content directly in their flow of work.

Azure Machine Learning is also adding support for foundation models allowing you to fine-tune and deploy foundation models from multiple open-source repositories using Azure Machine Learning components and pipelines. This feature will soon support models from Azure OpenAI Service as well.

The new Responsible AI dashboard support for text and image data will allow users to evaluate large models for errors, fairness issues and model explanations during the model building, training and/or evaluation stage.

The Model monitoring feature will allow users to track model performance in production, receive timely alerts and analyze issues for continuous learning and model improvement.

Additional updates supporting collaboration, governance and rapid development at scale for enterprise customers include:

  • Managed feature store, now in preview, will simplify feature development to streamline the machine learning lifecycle, helping users experiment and ship models faster, increase reliability of models and reduce operational costs.
  • Microsoft Purview connector, now in preview, will enable enterprises to use the Purview catalog as central storage for metadata for machine learning assets, allowing data engineers, data scientists and app developers to examine the lineage and transformations of training data and perform root cause analyses.
  • Managed network isolation, now in preview, will streamline complex tasks, such as virtual network management, private endpoint connections and inbound/outbound settings, saving users time and providing a more secure environment for machine learning projects.
  • Support for DataRobot 9.0, now in preview, will enable users to validate and document models in DataRobot and deploy with the scale and flexibility of Azure Machine Learning. DataRobot 9.0 is also integrated with Azure OpenAI Service, enabling code generation and conversational AI experiences to help users interpret model results.
  • Azure Machine Learning registries, now generally available, help users promote, share and discover machine learning artifacts such as models, pipelines and environments across multiple workspaces in an organization for more efficient cross-team operations and collaboration.
  • Azure Container for PyTorch, now generally available, provides users with a curated environment that includes the latest PyTorch 2.0 capabilities and optimization software such as DeepSpeed and ONNX Runtime designed for efficient large model training and inference.