Spotlight on AI at Google Cloud Next ‘18
Artificial intelligence has become a sort of secret weapon in the battle to build the best cloud service platform. Google Cloud Platform is currently the underdog, trailing both Amazon Web Services and Microsoft Azure. But Google is betting robust AI will give it the edge it needs to catch up. At the annual Google Cloud Next conference which kicked off July 24 in San Francisco the company unveiled a series of AI-based product releases and enhancements for its analytics and machine learning tools, additional applications on G Suite, and new IoT products.
Earlier this week, Google parent company Alphabet reported its Q2 earnings, which were ahead of Wall Street’s expectations. The company’s “other revenue” category, which includes the Google Cloud business, rose 37 percent over last year. CEO Sundar Pichai emphasized that “machine learning has been a major focus and a key differentiator for Google, and that’s true for our Google Cloud customers as well.”
Synced is onsite at the Moscone Center in San Francisco to bring you Google Cloud Next ’18 news and updates.
New AutoML tools: Natural Language and Translation
AutoML is one of Google Cloud’s standout innovations: A suite of machine learning tools designed for developers who have limited machine learning expertise and resources. Powered by neural architecture search and transfer learning, AutoML can help developers build custom AI models, or more specifically, find the best neural network architecture and automatically set weights.
Earlier this year Google Cloud announced AutoML Vision, which provides pre-trained models via an API, and the ability to build vision-based custom models. Google has now introduced AutoML Natural Language and AutoML Translation, which enable model customization and integrate with existing Google Cloud APIs. All three AutoML tools are currently available in beta.
To build a custom model, developers need only feed labeled text data into AutoML Natural Language or translated language pairs into AutoML Translation. AutoML will then produce a trained and optimized machine learning model.
Google provides labeling and annotation services, a critical data processing step which ensures training data is high-quality. AutoML Vision’s in-house human team manually classifies images with labels. Natural Language API can analyze text data, such as content classification, sentiment analysis, and entity recognition. Translation API can directly convert texts in more than 100 languages.
Cloud TPU V3 available for alpha users
The Tensor Processing Unit (TPU), introduced two years ago by Google, is a custom application-specific integrated circuit (ASIC) tailored for machine learning workloads on TensorFlow. In February Google made its TPU V2.0, or Cloud TPU, available in beta for researchers and developers on the Google Cloud Platform. Built with four custom ASICs, Cloud TPU delivers a robust 64 GB of high-bandwidth memory and 180 TFLOPS of performance.
At Google Cloud Next the company announced that its TPU 3.0 — a next-generation AI chip announced two months ago that is eight times more powerful than its predecessor and can achieve up to 100 petaflops performance — are now available in alpha. Google Cloud Chief AI Scientist Fei-fei Li comments, “TPUs allowed eBay to reduce the training time of their visual search model by a factor of almost 100 — from months to days.”
Contact Center AI, a virtual customer service representative
Google is applying its conversational AI technologies in customer service. The company announced Contact Center AI, a cloud-based machine operator that can answer customers’ questions, fulfill basic tasks, and transfer the calls to a human representative if necessary.
The Contact Center AI adopts some of the technologies behind Duplex, an advanced system announced two months ago that can for example call restaurants for reservations or schedule appointments at a hair salon.
Contact Center AI also leverages capabilities from Dialogflow, a technology that enables conversational AI and helps developers build chatbots. Google announced new Dialogflow features at Google Cloud Next, including text-to-speech capability via DeepMind’s WaveNet, the Dialogflow Phone Gateway for telephony integration, Knowledge Connectors to enrich the response database, and Automatic Spelling Correction.
Smarter Gmail, Hangout, and Google Docs available for G Suite customers.
Today, over four million businesses subscribe to Google G Suite, a set of cloud-based productivity and collaboration tools that includes G-mail, Google Calendar, Hangout, Google Drive, etc. Google is now using AI to promote its cloud-based enterprise businesses to the next level.
Google also announced that Smart Compose for Gmail will be soon available to G Suite customers. The program can automatically complete emails by filling in greetings or signing off, etc. It was first introduced with Gmail’s redesign this May as an experimental access feature.
Hangout, a Google communication platform that enables messaging and video chatting, added the new feature Smart Reply to help users respond to messages more quickly. Users are given automatic response suggestions on a Hangout Chat interface.
But the most exciting new feature might be the AI-powered grammar suggestions on Google Docs. Using machine learning, Google Docs can perform corrections from “simple grammatical rules like how to use articles in a sentence (‘a’ versus ‘an’), to more complicated grammatical concepts such as how to use subordinate clauses correctly.” The new feature was made available this week in Google’s Early Adopter Program.
Other announcements include a security center investigation tool (available in the Early Adopter Program for G Suite Enterprise customers), data regions (available now for G Suite Business and Enterprise customers), and voice commands in Hangouts Meet hardware (coming to select Hangouts Meet hardware customers later this year).
Cloud IoT Edge and Edge TPU
The Cloud and the Internet of Things (IoT) are inseparable, and Google has already invested heavily in IoT with products such as Android Things, Nest, Google Home, etc. Several months ago, the company announced Cloud IoT Core, a service that connects data from millions of dispersed devices using the Google Cloud Platform, and provides data-intensive processing, visualization, and analysis.
Without edge computing however, the back-and-forth communication between devices and the Google Cloud Platform would still cause high latency. Google took a step to fill that void yesterday with two new products: Edge TPU, and Cloud IoT Edge.
Edge TPU is a cut-down Google ASIC designed to complement Cloud TPUs, and will be embedded into gateways that bridge the Google Cloud Platform and devices such as sensors. Edge TPU empowers TensorFlow Lite machine learning models and accelerates inference at the edge so that edge devices can make local, real-time, intelligent decisions.
Cloud IoT Edge is a software that allows edge devices to run pre-trained machine learning models on the Edge TPU or on GPU- and CPU-based accelerators. Cloud IoT Edge can run on Android Things or Linux OS-based devices, and its runtime environment should include gateway class devices, the Edge IoT Core, and the TensorFlow Lite.
CTO of LG CNS Shingyoon Hyun says Google Cloud AI and IoT technologies will allow the South Korean tech company to “make a better working place, raise the quality of products, and save millions of dollars each year.”
Google will release an Edge TPU development kit, including a system on module (SOM) that combines Google’s Edge TPU, a NXP CPU, Wi-Fi, and other microchip’s elements Google has teamed up with semiconductor companies like NXP and Arm as well as gateway vendors and edge computing companies for manufacturing. The development kit will be available for order this October.
Most new products announced at Google Cloud Next ’18 so far are based on existing Google technologies such as AutoML, TPU, and other AI features. They are perfectly integrated with Google cloud business and designed to solve the pain points of businesses.
Google Cloud’s quarterly revenue crossed a billion dollars in 2017, prompting Google Cloud CEO Diane Greene to proclaim it the world’s fastest growing cloud. There is nothing to suggest that growth will slow in the future, as the cloud plays an important role in Google’s grand strategy to transition from an Internet giant into an AI mega-power.
Journalist: Tony Peng | Editor: Michael Sarazen
Follow us on Twitter @Synced_Global for more AI updates!
Subscribe to Synced Global AI Weekly to get insightful tech news, reviews and analysis! Click here !
