Google launches an end-to-end AI platform – TechCrunch

As expected, Google spent the second day in its annual Cloud Next conference shining a spotlight on its AI tools. The company made a staggering number of announcements today, but at the heart of all these new tools and services is the company's plan to democratize AI and machine learning with pre-built models and easier to use services while providing more advanced developers with the tools to build their own. custom models.

The highlight of today's announcements is the beta launch of the company's AI platform. The idea here is to offer developers and computer scientists an end-to-end service for the construction, testing and distribution of their own models. To do this, the service gathers a number of existing and new products that allow you to build a full data pipeline to retrieve data, mark it (using a new built-in branding service), and then use existing classification, object recognition or device extraction models, or Use existing tools such as AutoML or the Cloud Machine Learning engine to train and deploy custom models.

"The AI ​​platform is this place where if you take this scary journey from a travel idea of ​​how to use AI in your business, through launch and secure and reliable deployment, the AI ​​platform helps you move between each of these steps safely so that you can begin exploratory data analysis, start building models using your computer science, decide that you want to use this specific model, and then distribute it with essentially one click said a Google spokesman during a press conference before today's official announcement. 659002] But there is also much more AI news, mostly courtesy of Cloud AutoML, Google's tool for automating the model training process for developers with limited machine learning skills. [19659002] One of these new features is AutoML Tables, which takes existing table data that can sit on the Google BigQuery database or in a storage service, and automatically creates a model that will predict the value of a given column.

Also new is AutoML Video Intelligence (now in beta), which can automatically annotate and capture video, using object recognition to classify video content and make it searchable. To discover objects in images, Google today also launched the beta of AutoML Vision, and for edge-edge applications, Google launched Beta AutoML Vision Edge, which includes the ability to deploy these models to edge units.

Many corporate data come in the form of fair, unstructured text, though. For these applications, Google today launched the beta of its custom device recovery service and a customized sentiment analysis service. Both of these tools can be adapted to the needs of a given organization. It is one thing to use a generic recovery service to understand documents, but for most businesses, the real value here is to be able to extract information that can be very specific to their needs and processes.

Talking about documents, Google has also today announced the beta of its document understanding API. This is a new platform that can automatically analyze scanned or digital documents. The service basically combines the ability to scan a scanned page into machine-readable text, and then use Google's other machine learning services to extract data from it.

Having introduced it in preview last year, the company today launched the beta of its Contact Center AI. Built with partners such as Twilio, Vonage, Cisco, Five9, Genesys and Mitel, this service offers a full AI contact area solution that uses tools such as Dialogflow and Google's text-to-speech capabilities to allow users to build a virtual agent system (and when it goes wrong, it can pass the customer to a human agent).

It's no secret that many companies struggle to combine all these tools and services into a coherent platform for their own needs. Perhaps it's not surprising when Google also launched the first AI solution for a particular vertical: Google Cloud Retail. This service combines the company's Vision Product Search, Recommendations AI and AutoML Tables in a single solution to cope with small-scale use. Chances are, we'll see more of the packages for other verticals in the near future.