Google's main enterprise-focused event – Google Cloud Next – may have already been and gone, but there were a number of announcements at its I/O developer conference yesterday that furthered the company's aims of pushing machine learning to large businesses.
Google I/O 2017 announcements: Machine learning
Opening the two-hour keynote, CEO Sundar Pichai announced the second generation of its Tensor Processing Units – custom-made processors optimised for machine learning workloads.
The latest version of the chips will enable users to both train and run machine learning models, the company said in a blog post, and can be combined into 'TPU pods' which will increase the computational capacity and speed training times. Google claims that the TPUs are much faster than competing commercially available GPUs for similar tasks.
The TPUs will soon be available to businesses via its Google Compute Engine public cloud service, which has grown in popularity among corporate customers over the past few years.
As cloud adoption begins to mature, with basic compute and storage feature becoming commodified to some degree, the battle between the big providers is moving to higher-level services such as machine learning, and offering access to Google's internal AI expertise is a major selling point.
Google's TPUs were, for example, used to support DeepMind's AlphaGo project, and power the machine learning behind well-known services such as Google Photos. It is not the only cloud provider to offer such services, however, with Amazon Web Services and Microsoft Azure providing access to GPUs via their respective cloud platforms.
In order to encourage use among academic researchers, Google announced that it will make 1,000 Cloud TPUs available for free as part of its TensforFlow Reseach Cloud initiative.
Google has already been working to apply machine learning to the healthcare sector via its UK-based DeepMind division, and these aims were furthered with the announcement of a partnership with medical researchers in the US. The company says that it is working with staff at UC San Francisco, Stanford Medicine and Univerisity of Chicago Medicine to explore how machine learning can be combined with clinical expertise to improve patient outcomes.
Also announced by Pichai was Google.ai – a new initiative that aims to make its machine learning available to a wider audience. It will host research from within Google, as well as from researchers at universities.
Google I/O 2017 announcements: VR
As expected, virtual reality was another key focus of the opening keynote, with a new standalone headset unveiled. This builds on the DayDream VR announcement at last year's I/O event – and the launch of its DayDream View headset last October – which required a smartphone to run the software. The new hardware will be produced by Lenovo as well as HTC, which has had success with its Vive headset for PCs.
While DayDream was available on only a limited selection of smartphones, Google also announced that it will be supported by Samsung's flagship Galaxy S8 handset, which will likely give the VR platform a significant boost in terms of adoption.
Despite the huge industry hype around VR it is still unclear when – or if – it will become a real mainstream proposition that businesses need to develop for in a more meaningful way. However, the announcements from Google and its partners are another step in this direction.
Google I/O 2017 announcements: IoT
Google also unveiled a new internet of things service – Cloud IoT Core – the day before I/O kicked off. Running on Google Cloud, the service will manage IoT devices and help developers incorporate analytics data into applications. Aimed at industrial firms such as utilities and manufacturing companies, Google says that Cloud IoT Core simplifies the process of setting up IoT pilots, as well as scaling projects up.
Read on for enterprise highlights from last year's Google I/O conference
Google I/O 2016 announcements: Daydream VR and Android Wear 2.0
Google stopped short of unveiling a new VR headset, as many had predicted, but one of the most eye-catching announcements was Daydream. Daydream was billed as the VR platform for Android. Google is working with Android manufacturers to create devices that can run a high-end VR experience.
There were also plans to build on the success of Cardboard with reference designs for a Daydream headset, alongside a controller similar to that of the HTC Vive.
Considering the reach Google has with its Android platform, the announcements were expected to give VR adoption a significant boost in the next year.
Forrester analyst J.P. Gownder commented: "From what we saw in the brief I/O presentation, our initial belief is that Daydream will play a critical role in driving forward the VR ecosystem for both enterprises and consumers.”
This is partly because Cardboard doesn't have a great experience, while mobile-centric hardware has more “mass-market reach”.
While this was good news for consumers, it also offered a new channel of interaction for businesses.
“Broader offerings from more handset makers will offer both consumers and enterprises opportunities to leverage VR,” Gownder said.
“Six Flags is deploying Gear VR devices on its rollercoasters, and real estate agents are experimenting with showing properties on the smartphone-based VR platform."
You can find out more about Daydream on PC Advisor: Google Android Daydream VR release date and features
Google also updated its Android Wear platform, with a developer preview of Wear 2.0 - the most significant upgrade since launching two years previously. New features included the ability to access the internet directly from Android Wear, meaning that wearable devices would function even if your phone was not nearby, or out of battery, as well as a new UI and input methods.
Google I/O 2016 announcements: Firebase
One announcement that was certainly welcomed by developers was the Firebase update. Acquired by Google in 2014, Firebase makes it easier to build apps and was expanded significantly at last year's conference, turning it into a unified app platform with support for iOS, Android and web development.
New features included an integrated analytics tool – which Google billed as 'Google Analytics for developers' - providing data on how applications are being used by an audience or to view advertising performance.
There is also a new notification system based on Google Cloud Messaging, and two new offerings to improve app quality – Test Lab and Cash Reporting.
Google I/O 2016 announcements: Machine learning - Tensor Processing Units
One of the most interesting announcements was the unveiling of its Tensor Processing Units. TPUs are specialised chips tailored to machine learning applications – and specifically for its TensorFlow framework.
CEO Sundar Pichai made some intriguing claims, with TPUs apparently achieving ten times better performance than comparative hardware such as GPUs and FGPAs for machine learning purposes. This is equivalent to fast-forwarding three generations of Moore's Law, he said.
Google hasd been developing the custom ASICs [application specific integrated circuit] in secret for a number of years, with TPUs running in their datacentres for more than a year at that point, powering applications such as RankBrain. The chips were even used by AlphaGo in the matches against world champion Lee Sedol.
Google was clearly betting big on machine learning. As well as TensorFlow, which was open sourced the previous year to encourage community development, it had been opening up some of the APIs used in its own products. Pichai stated that these APIs are "one of the biggest differentiators for Google Cloud Platform" as it aimed to draw developers away from AWS and towards its own services.
Find your next job with computerworld UK jobs