Google Uses DeepMind to Run Its Data Centers Efficiently; And Other News


Google Uses DeepMind to Run Its Data Centers Efficiently

DeepMind famously came to the fore when its AI, AlphaGo, beat Lee Sedol in a series of Go matches, the first time in which a program beat a human professional at the notoriously complex game.

Now it appears that the advanced machine learning capabilities provided by DeepMind are providing tangible benefits for Google, which owns the company. Google has announced that it is ceding control of all of its data center cooling operations to DeepMind, after an initial trial run. The trial used DeepMind’s AI to manage the cooling of some of its data centers, which led to a 15% improvement in power usage efficiency, which helps reduce costs as well as the energy-footprint data centers.

Data centers are huge cost and energy sucks, accounting for a sizeable percentage of the 4,402,836 MWH of power that Google uses each year, according to Bloomberg. Given the high cost of energy, DeepMind can save Google hundreds of millions of dollars in utility bills over a span of just a few years.

DeepMind’s AI technology uses neural networks to analyze power usage and predict spikes in energy needs over time, thereby calibrating the equipment to maximize efficiency.

“It controls about 120 variables in the data centers. The fans and the cooling systems and so on, and windows and other things,” DeepMind CEO Demis Hassabis said to Bloomberg. “They were pretty astounded.”

Verizon Prepares for NG-PON2 Testing

Verizon has announced that by the end of this month, it will trial its landmark fibre-optic technology, NG-PON2 (next generation passive optical network), which can reach blazing-fast symmetrical speeds of up to 10 Gbps over FiOS. The NG-PON equipment is aimed at FiOS enterprise and consumer based services and was developed via its partnership with Ericsson, Calixx and Adtran. The trial is slated to occur in Verizon’s Innovation Lab located in Waltham, Massachusetts.

The technology was deemed acceptable by the International Telecommunications Union Telecommunications Standardization Sector and is expected to pave the way for Verizon’s most significant wireline telecommunications network upgrade in nearly a decade. What sets NG-PON2 apart is that it relies on four different wavelengths, which are denoted by different colors, to transfer data with a total capacity of 40GB per fiber. It also allows Verizon to deliver different services over different wavelength on the same fiber, differentiating between its residential and enterprise consumers.

Enterprise customers will be the first to benefit from the fruits of the testing as early as next year, if it proves to be successful, with consumers and residential clients gaining access to it later on.

The first successful field trial of NG-PON2 tech occurred last year in Framingham, Massachusetts.

Verizon is touting its fiber network as highly scalable and being easily upgraded: ““By implementing this advanced technology without having to change the current underlying fiber optic infrastructure, additional traffic can be carried cost effectively,” the company noted in a statement. “Verizon also can improve flexibility and resiliency using NG-PON2, because traffic can be shifted amongst multiple wavelengths without impacting customers.”

Huawei is Partnering with GE for IoT and Smart Cities in China

GE is bringing its IoT expertise to China, according to a new press release announcing its partnership with Chinese telco manufacturer Huawei.

The primary focus of its collaboration will be to provide smart machines that boost productivity and enhance worker output, which is a subset of IoT known as the industrial internet of things. The enticing promise of IIoT is that it combines big data analytics with machine learning, driving massive gains in productivity.

GE is investing heavily in an $11 million incubator, which will work on developing start-ups and software to improve machine intelligence, according to Reuters. The digital platform that developers will build their applications on is called Predix, which is owned by GE. Such applications would work with connected machinery to collect and analyze data in real-time.

GE has recently undergone a massive restructuring and shifted its focus to IoT, having invested more than $500 million in software annually.

In a recent statement to shareholders, GE noted: “We are a company that invests in broad industrial transitions, and they don’t come much bigger than the full application of data and analytics to machines and systems.”

The move has paid dividends; GE is expected to make $6 billion in revenues this year. Chief Digital Officer Bill Ruh notes that using smart machines had saved the company more than $500 million in production costs and predicts that such savings will increase to over $1 billion by 2020. This experience has motivated GE to bring its smart machine solutions to other partners and countries.

“Once we got it right for ourselves we take it to our customers … We’re bringing this to China, we’re open for business in China today to be able to do this,” he said.

GE’s business strategy of trialing solutions internally and translating them into saleable services has been working. Predix is not only used for internal manufacturing equipment and has been deployed to create new revenue streams for clients as well.

Such developments could also be used to innovate in the smart city space, which is a massive initiative that is being undertaken in China. Huawei reports that there are more than 300 smart cities planned in the nation, which necessitates collaboration with third-parties. GE can also bring its proprietary smart-city solutions suite, called Current, to the mix. Current leverages a network of LED technology, solar panels, and wireless controls to create green urban spaces, allowing cities to save on utility costs by generating power rather than purchasing it.

Google Expands Cloud Platform to Western Region of North America

Google has announced in a blog post that it is extending its cloud platform to customers in the west coast of North America, dubbing the region the Oregon Cloud Region. West Coast customers based in Vancouver, Seattle, Portland, San Francisco and Los Angeles can expect to see a 30-80% latency reduction. They also now have access to the Google Compute Engine, Cloud Storage, and Container Engine.

Latency reduction is crucial to providing consistent and immersive gaming experiences and is one of the biggest issues encountered by gamers. As Google client Multiplay, a video game hosting specialist, notes:

“Regional latency is a major factor in the gaming experience. Google Cloud Platform’s network is one of the best we’ve worked with, from a tech perspective but also in terms of the one-on-one support we’ve received from the team.”

As we noted in an earlier blog post, Google has also invested in a massive trans-Pacific cable network in furtherance of its aim to establish a Cloud Platform region in Tokyo. The Japan-based region is expected to commence operations later this year.

Google Unveils Two New Cloud Machine Learning API’s

Google has released a blog post announcing two new Cloud Machine Learning products– Cloud Natural Language and Cloud Speech API’s.

The Cloud Natural Language API is the fruit of its efforts to improve machine learning and acquisition of human language understanding, so that computers can parse the intricacies of syntax more accurately. It initially supports English, Spanish, and Japanese and can assess the general sentiment being expressed in a statement, identify the various entities mentioned (be they companies, persons, media, or animals), and parse the structure of a sentence.

Google believes that its natural language acquisition API can utilized to great effect in a wide range of industries: “ For example, digital marketers can analyze online product reviews or service centers can determine sentiment from transcribed customer calls.”

The Cloud Speech API serves a related but different function, specializing in speech-to-text conversion for over 80 different languages, applying voice recognition technology to use-cases in apps and IoT. It leverages and applies “advanced deep learning neural network algorithms to [user’s] audio for speech recognition with unparalleled accuracy. Speech API accuracy improves over time as Google improves the internal speech recognition technology used by Google products.”

The API allows machines to take advantage of tailored word hints to optimize recognition of certain relevant phrases, words, or neologisms. For instance, Google notes that a smart TV would be listen for keywords such as “fast forward.” Cloud Speech also features asynchronous calling that allows for faster and easier app development and has been fine-tuned to function in noisy environments.

Scroll to Top