Microsoft Corp. is pushing to become a more important player in artificial intelligence development, leveraging its partnerships and expertise to offer customers tools for building more advanced models.

One of the most impressive new offerings comes from Microsoft’s Azure OpenAI Service. Falling under the umbrella of Azure Cognitive Services, it provides access to an array of powerful models from the artificial intelligence research outfit OpenAI Inc.

Announced today at Microsoft Ignite 2022, Azure OpenAI Service users will now be able to access OpenAI’s newest DALL-E 2 offering, which is an advanced AI model that can be used to generate stunning visual images using either text, or more basic pictures as input. It’s a powerful natural language-to-image generator.

DALL-E 2 has been difficult to access until recently, so its availability on Azure OpenAI Service should ensure it will be used to good effect. Based on the original DALL-E tool, it provides new capabilities such as being able to edit existing pictures. When first announced, OpenAI showed how the model can be used to create depictions of almost anything, ranging from mundane mannequins to flannel shirts and even abstract ideas, such as a “giraffe made of turtle.”

DALL-E 2 also comes with a new “inpainting” feature that applies its text-to-image prowess on a more granular level. So, for example, users can start with an existing picture, choose just a part of it, and edit that section. The user could ask DALL-E 2, say, to block out a picture hanging on a wall and replace it with something new, or else remove or add objects on a table.

The model is so precise that it will even take into account how adding or removing something affects other details in the picture, such as light and shadows. Add a candle to a table and it will automatically generate a halo-like effect around it, showing any details nearby in brighter light.

Microsoft said the addition of DALL-E 2 will expand the breadth of use cases for its Azure OpenAI Service. It’s currently available in limited access in preview.

AI service enhancements

There were plenty of other AI updates in store at Ignite. The Azure Cognitive Service for Language is said to enhance summarization and expand language support across language skills with more natural language understanding and generation capabilities, so customers can efficiently implement business apps for document and conversation digitalization scenarios.

Microsoft is introducing expanded summarization capabilities in preview, such as abstractive document and meeting summarization tools, complete with chapter segmentation. There’s also increased language coverage, with support for more than 90 languages in total, plus enhanced contact center functionality.

Microsoft Azure’s Computer Vision offering gains two new services, with both Image Analysis 4.0 and Spatial Analysis on the Edge now available in preview. The first is an update to a service that’s designed to extract a wide variety of visual features from images to improve digital asset management and customer accessibility. The second is meant to improve safety and security by ingesting streaming video from cameras, pulling out insights and then generating events that can be used by other systems, for example warning notifications.

In the same vein, Azure Cognitive Service for Speech gains new capabilities around speech-to-text and text-to-speech that will benefit media experiences, customer service and accessibility. For instance, Neural Text to Speech now supports additional languages and emotions, helping to improve the ability of AI-based voice assistants. Moreover, Custom Neural Voice gets synthetic voice training, including multi-style tones such as cheerful or sad. And Embedded Speech is an entirely new capability that makes it possible to use Azure Speech Services on company devices.

Azure Machine Learning

On the AI developer side, Azure Machine Learning has gained new features such as Machine Learning Registries, which make it possible for developers to promote, share and discover artifacts such as models, pipelines and development environments across multiple workspaces. With this update, teams will be able to track model and data lineage across various workspaces, enabling superior collaboration and cross-team operations.

Available in preview starting today are the new Azure Container for PyTorch and Azure Data Science Virtual Machines for PyTorch curated environments and custom images. According to Microsoft, these bundle various technologies for setting up, developing, accelerating and supporting optimized training for PyTorch models.

Also coming soon is a new Responsible AI Dashboard within Azure Machine Learning. The idea with this is that developers can easily implement responsible AI by debugging their ML models and making better informed, data-driven decisions. One of its key components is a Responsible AI Scorecard, which provides a way for low-code and no-code users to create reports about their ML models.

Project Bonsai

Microsoft’s Project Bonsai is a low-code industrial AI development platform currently available in preview that allows engineers to create AI-powered automations with minimal coding skills. It too is getting a host of updates that should make life more enjoyable for users.

One of the most interesting new features in Bonsai is the addition of support for action masking, which helps to guarantee valid behavior and speed up model training by restricting the available actions an AI system can take, based on the current state of the system. As an example, Microsoft said it’s now much easier to train a system to understand that it cannot route work to a factory machine that’s currently undergoing maintenance.

Also new is OfflineRL, which enables users to build systems that can control or optimize processes using a dataset of states and actions from a historian, “internet of things” system or other record. Then there’s the new model-based training acceleration feature, which introduces the ability to accelerate training for slower simulations. Microsoft said this will reduce the time and computational costs of training control policies based on a specific customer use case.

Other new features include Assessment 2.0, which provides richer analytical tools for evaluating automation performance, and the availability of Bonsai Samples in the Azure Marketplace. Bonsai Samples are examples and simulations of models that citizen developers can use to get started with building their own automations.

Finally, Microsoft said Project Bonsai is getting a new partner in Tata Consultancy Services, which will help customers to build intelligent industrial control systems and AI-powered automations for very specific use cases.

Image: OpenAI