It could not be otherwise: at Google Cloud Next 19 artificial intelligence and machine learning technologies are among the main protagonists. Rajen Sheth, Director of Product Management of Google Cloud AI, explains what are the most important innovations in this area.
And there are a lot of new ones. Starting with the introduction of a new integrated service platform that helps develop artificial intelligence capabilities and then execute them in the cloud or on-premise. So let’s start with this one, which is not the only one.
Google AI Platform, for teamwork
When they approach artificial intelligence projects, Rajen Sheth explains, companies face a number of problems. From unstructured data to silos teams to complexity of implementations. According to Goole, companies need a place that brings together all these elements in a way that makes machine learning easier and more collaborative.
Google’s response to this set of problems comes with the announcement of the AI Platform, currently in beta version. It is a complete end-to-end development platform. Designed to help teams prepare, create, run and manage machine learning projects through the same shared interface. AI Platform is aimed at developers, data scientists, data engineers, and everyone can collaborate in sharing models, training and scaling workloads. All of it, from the same dashboard inside Cloud Console.
Ai Platform, explains Google, makes it possible to import data from streaming or batching. And use an integrated labeling service to easily label training data such as images, videos, audios and text. This, by applying classification, object detection, entity extraction and other processes.
The platform allows you to import data directly into AutoML or use Cloud Machine Learning Engine, now part of AI Platform. In order to train and serve on GCP its own custom machine learning models. AI Platform integrates AI Hub. And since it supports Kubeflow, Google’s open source platform, you can create portable machine learning pipeline. Kubeflow pipeline that you can then run on-premise or in the cloud, almost without any code modification.
More information about the AI Platform is available on the Google website at this link.
For a more accessible artificial intelligence
Google introduced AutoML Cloud for the first time in early last year. The company’s goal was to help developers with limited machine learning skills in training high quality custom machine learning models. And in their implementation.
At Google Cloud Next 19 the company announced a series of new and improved AutoML solutions. New things that aim to make it even easier and faster for developers and businesses to use artificial intelligence.
AutoML Tables, available in beta version, enable build and deploy of state-of-the-art machine learning models on structured tabular data sets, without any code. With just a few clicks, the tool allows you to import data from BigQuery and other GCP storage services into AutoML tables. To create and distribute machine learning models in a few days instead of weeks.
The interface guides the user, without the use of any code, through the entire life cycle of machine learning. Making it easy for anyone in the team, whether data scientist, analyst or developer, to create models and incorporate them into wider applications.
More information about AutoML Tables is available at this link.
Other innovations of AutoML and the entire infrastructure
Optimization of machine learning models for execution on edge devices, such as sensors or connected cameras, can prove difficult. This is because these devices are often subject to unreliable latency and connectivity. AutoML Vision was introduced last year with the aim of simplifying the creation of custom models for image recognition.
On the occasion of Google Cloud Next 19 Google announced AutoML Vision Edge to simplify the training and deployment process of custom machine learning models with high precision and low latency for AutoML Vision Edge supports a variety of devices and can leverage the Edge TPU for faster inference.
AutoML Video, currently in beta version, allows you to create custom models that automatically classify video content with defined labels. This allows companies that manage huge amounts of different video data to raise content instantly, based on their taxonomy.
In addition to these three completely new solutions, Google has continued to improve the basic features of AutoML Vision and AutoML Natural Language.
In addition, the company continues with the upgrading of the infrastructure supporting these artificial intelligence and machine learning services. Third-generation liquid-cooled TPUs are now generally available. In addition, all TPU Clouds are generally available in Google Kubernetes Engine (GKE). GCP is also the first cloud service provider to offer the new NVIDIA Tesla T4, now generally available in eight regions.
Google also announced its ongoing collaboration with numerous partners, including Accenture, Atos, Cisco, Gigster, Intel, NVIDIA, Pluto 7, SpringML and UiPath. A collaboration aimed at building Kubeflow pipeline to grow and extend AI Hub.