Source: Expanding Google Cloud AI to make it easier for developers to build and deploy AI from Google Cloud
Every year, more and more businesses look to AI to help them solve complex business challenges. Whether they’re using AI to anticipate demand, predict when equipment will need routine maintenance, or deliver better customer experiences, they all have one thing in common: they need a workforce that can help them do it.
Our goal has always been to make AI simpler, faster, and more useful for businesses. This means easy-to-use AI solutions that make it simple for enterprises to adopt them. But it also means making it simpler for developers, data scientists, and data engineers to build and deploy machine learning models.
Today we’re announcing a number of new ways we’re doing exactly that—from introducing an integrated platform of AI services that helps you build AI capabilities, then run them in the cloud or on premises, to expanding our AutoML offerings to make it easier for businesses to build and deploy their own custom ML models.
Here’s a selection of what’s new:
AI Platform (beta)
AutoML updates, including:
AutoML Tables (beta)
AutoML Video Intelligence (beta)
AutoML Vision Edge (beta)
Object detection (beta)
AutoML Natural Language
Custom entity extraction (beta)
Custom sentiment analysis (beta)
When approaching AI projects, businesses grapple with a variety of problems—from unstructured data to siloed teams to complex deployments. They need a place that brings all these things together in a way that makes ML easier and more collaborative.
Today, we’re announcing AI Platform in beta, a comprehensive, end-to-end development platform that helps teams prepare, build, run, and manage ML projects via the same shared interface. Whether you’re a developer, data scientist, or data engineer, you can collaborate on model sharing, training, and scaling workloads from the same dashboard within Cloud Console.
With AI Platform, you can ingest streaming or batch data, and use a built-in labeling service to easily label training data—like images, videos, audio, and text—by applying classification, object detection, entity extraction, and other processes. You can import your data directly into AutoML, or use Cloud Machine Learning Engine, now part of AI Platform, to train and serve your own custom-built ML models on GCP. AI Platform complements AI Hub, so developers can discover ML pipelines, notebooks, and other instructional content, and because AI Platform supports Kubeflow, Google’s open-source platform, you can build portable ML pipelines that you can then run on premises or in the cloud with almost no code changes.
Learn more about AI Platform on our website.
When we first introduced Cloud AutoML, our goal was to help developers with limited ML expertise train high-quality custom machine learning models and deploy them in their business. Today, we’re excited to announce new and enhanced AutoML solutions that will further our mission of making it easy, fast, and useful for all developers and enterprises to use AI.
AutoML Tables: easily create ML models from datasets with no coding necessary
Enterprises are generating more structured data than ever, and tools that help them easily turn all that data into actionable predictive insights can be a huge help. AutoML Tables, now available in beta, lets you build and deploy state-of-the-art machine learning models on structured tabular datasets with zero code. With just a few clicks, you can ingest data from BigQuery and other GCP storage services into AutoML Tables and build and deploy ML models in just days versus weeks. The codeless interface guides you through the full end-to-end machine learning lifecycle, making it easy for anyone on your team—whether data scientist, analyst, or developer—to build models and reliably incorporate them into broader applications.
For an ever deeper look at AutoML Tables, read our data analytics blog post.
Extending AutoML Vision to the edge
Optimizing machine learning models to run on edge devices, like connected sensors or cameras, can be challenging because these devices often grapple with latency and unreliable connectivity. Last year, we announced AutoML Vision to make it easier for developers to create custom ML models for image recognition. Today we’re announcing AutoML Vision Edge to simplify training and deployment of high-accuracy, low-latency custom ML models for (on premises or remote) edge devices. AutoML Vision Edge supports a variety of devices and can take advantage of Edge TPUs for faster inference. For example, LG CNS is using AutoML Vision Edge to create manufacturing intelligence solutions that detect defects in everything from LCD screens to optical films to automotive fabrics on the assembly line.
Enabling powerful content discovery and engaging experiences with AutoML Video
Analyzing volumes of video footage to identify specific moments, prepare special cuts, or better classify visual data can be a difficult and time-consuming process. Today, we’re announcing AutoML Video, in beta, so that developers can easily create custom models that automatically classify video content with labels they define. Companies that deal with mountains of diverse video data can instantly discover content according to their own taxonomy. This means media and entertainment businesses can simplify tasks like automatically removing commercials or creating highlight reels, and other industries can apply it to their own specific video analysis needs—for example, better understanding traffic patterns or overseeing manufacturing processes.
In addition to these three entirely new AutoML solutions, we are continuing to improve the core functionality of AutoML Vision and AutoML Natural Language. AutoML Vision object detection (beta) can identify the position of objects within an image, and in context with one another, for example, a pedestrian walking in a crosswalk. AutoML Natural Language custom entity extraction (beta) helps you automatically identify entities—such as medical terms or contractual clauses—within documents and label them based on company-specific keywords and phrases. And AutoML Natural Language custom sentiment analysis (beta) helps you apply machine learning to better understand the overall opinion, feeling or attitude expressed in a block of text, tuned to your organization’s own domain-specific sentiment scores.
We continue to invest in the infrastructure that makes machine learning possible for you. Our Cloud TPUs, custom-built to quickly train ML models, lets you iterate at scale to achieve higher classification accuracy, at a lower cost. Our third generation liquid-cooled TPUs are now generally available, and all Cloud TPUs are also generally available in Google Kubernetes Engine (GKE), which is a new and flexible way to run your containerized ML workloads, giving you the flexibility to switch between on-prem and cloud-based training. GCP is also the first cloud provider to offer the new NVIDIA Tesla T4, now generally available across eight regions.
As part of today’s announcements, we’re also working with numerous partners—including Accenture, Atos, Cisco, Gigster, Intel, NVIDIA, Pluto 7, SpringML, and UiPath—to build Kubeflow pipelines to grow and extend AI Hub. It takes a robust partner ecosystem to build a successful platform, and we’re grateful to all of our partners who enable our customers to train and serve machine learning pipelines on the infrastructure of their choosing.