Available AI for any company: Cloud AutoML

Hi, Habr! Today, information is shared by Jia Lee, head of R & D, Cloud AI. Gia and the team made the AI ​​easy to use and accessible even to non-specialists. We hope that now AI will come to every business, as the computer once came to every home, and we read what Cloud AutoML looks like and what it can do.

When Fay-Fay and I (Cloud AI, Chief Research Officer) joined Google Cloud a year ago, our goal was to make AI available to as many developers, researchers, entrepreneurs, and businesses as possible. Our Google Cloud AI team has done a lot to achieve this goal. In 2017, we introduced the Google Cloud Machine Learning Engine to help developers in this area create models that can work with data of any type and size. We showed how modern machine learning services, for example, various APIs , including Vision, Speech, NLP, Translation and DialogFlow, created on the basis of pre-trained models, can help scale up and speed up business applications. Kaggle , our data processing and machine learning community, has grown to over a million participants. And now Google Cloud services, which use AI, are used by more than 10,000 companies, including Box , Rolls Royce Marine , Kewpie and Ocado .

But we are capable of more. Now only a small number of companies in the world have enough employees and a budget to engage in the development of machine learning and AI. Very few can create advanced machine learning models. And even if your company works with engineers in the field of AI and ML, you still need to build a time-consuming process to develop your own custom machine learning model. And, although Google already offers an API based on pre-trained models to perform certain tasks, much remains to be done to make AI available to everyone.

To fill this gap, we present Cloud AutoML - a product that will make AI available to any business. Thanks to Cloud AutoML, businesses with little expertise in machine learning have the opportunity to create their own unique models using advanced Google techniques, such as learning2learn and transfer training . We hope that thanks to Cloud AutoML, specialists in the field of AI will be able to work more efficiently, master new areas of application of their knowledge and help less experienced developers create their own powerful systems on the basis of AI, which previously could only be dreamed of.

The first Cloud AutoML product we represent will be Cloud AutoML Vision . With it, you can use a simple drag-and-drop interface to upload images, train and manage models, and then use these models directly in Google Cloud. The first results showed a significant improvement in image recognition in comparison with previous models that we used before. You can request access to Cloud AutoML Vision through this form.

Advantages of Cloud AutoML Vision over conventional machine learning APIs:

• High accuracy. CloudAutoML Vision uses Google’s advanced image recognition techniques, including transfer training and search technologies using neural networks , which have been able to improve work efficiency and reduce the number of classification errors in such well-known open data sets as ImageNet, CIFAR, etc. you are not enough expertise in machine learning, to create a more accurate model will be faster than customizing it yourself.
• Quick calculation of ready-to-use model. With Cloud AutoML, you can create a simple model in an application that supports AI in a few minutes, or build a complete, ready-to-use model in less than a day.
• Ease of use. Using the simple AutoML Vision GUI, you can specify the data you need and then convert it into a quality model that fits your needs.

AutoMLVision is the first Cloud AutoML product to simplify the use of machine learning in various businesses. This is the result of close collaboration between the Google Brain and Machine Perception teams. And although we have just started working on the availability of AI, we are very inspired that more than 10,000 customers are already using Cloud AI products. We hope that Cloud AutoML will help a much larger number of companies to evaluate the capabilities of AI.

If you are interested in trying AutoML Vision, you can request access through a special form.