Microsoft is trying to simplify machine learning – TechCrunch

Before the conference on the construction, Microsoft today has released a series of new machine learning products and changes to be made to some of its existing services. These range from tools without code to hosted notebooks, with a number of new APIs and other services in between. The central theme, however, is that Microsoft is pursuing its strategy of democratizing access to AI.

Prior to publication, I met Eric Boyd of Microsoft, vice president of the company's AI platform, to discuss Microsoft's position on this space, where it competes heavily with Google, AWS as well as many others, often more. specialized startups. And to some extent, the actual machine learning technologies have become table issues. Everyone now offers pre-trained templates, open-source tools and platforms to train, build and deploy models. If a company does not have pre-trained templates for some of the use cases that its competitors take care of, it's just a matter of time. However, these are ancillary services and the overall developer experience, where companies such as Microsoft, with long experience in developing these tools, can differentiate themselves.

"Artificial intelligence really has an impact on how the world is evolving," Boyd said. "We're seeing 75% of commercial companies doing more with AI in the next few years. He has tripled in the past two years, according to Gartner. We are witnessing an explosion in the amount of work that results. As companies move forward, as businesses do, developers are on the front line trying to find a way to grow their business, build these models and build those applications, and help them evolve changes that move through this. "

What these companies – and their developers – need are more powerful tools that allow them to become more productive and build their models faster. At Microsoft, where these companies are often large companies, it also means being able to adapt to the needs of a company and offer the security guarantees they need.

However, as companies adopt machine learning, they are now at the point where they have gone from a few tests to a hundred models in production. It comes with its own challenges. "They are trying to understand how to manage the life cycle of these models," he said. "How do I think about the operational cycle? How can I think of a new model that I am ready to deploy? When is he ready to leave?

A few years ago, the industry began to adopt a DevOps model for code management. What Microsoft essentially wants to do is MLOps technology for model management. "It's very similar to DevOps, but there are some differences in how the tools work," Boyd said. "At Microsoft, we are really focusing on how to solve these problems in order to make developers more productive, using these business tools to drive the changes they need across their entire organization. and the continued development of machine learning models, for example, requiring new tools.

It is therefore not surprising that the addition of new MLOps features is an important part of the current versions. For example, the company incorporates some of these features into Azure DevOps, which allows them to trigger release pipelines. The company also provides developers and data experts with tools for model version control, for example, to track and manage their assets and to share machine-learning pipelines.

Yet they are tools for advanced practitioners in machine learning. On the other side of the spectrum, Microsoft has also announced a number of automated machine learning tools, including a tool that essentially automates all processes, as well as a visual template builder. , from the studio Azure ML. As Boyd told me, even companies like British Petroleum and Oregon's Deschutes Brewery (try their Black Butte Porter if you can) are now using these tools.

"We've added many features to automated machine learning to simplify the way people try to use this type of work," Boyd said.

Today, Microsoft has also launched a number of new services in its cognitive service offering, including a new customization service, an API for handwriting recognition and another for transcribing conversations. to several speakers. The personalization service is distinguished by the fact that it uses reinforcement learning, a machine learning technique that is different from most other Cognitive Services tools, and is much easier to implement than similar services. For business users, there is also the Form Recognizer, which makes it easy to extract data from forms.

What's more interesting than the specific features, though, is that Microsoft is changing its focus a bit here. "We are moving away from the top-level issues of" here are the issues at the table, you need to have an AI platform, "for much more sophisticated use cases around the operations of these algorithms, their simplification, the new user. experiments to really simplify the work of developers and much richer cognitive services, "explained Boyd.