Oracle puts GraphPipe into open source to standardize and deploy machine learning

machine learning
Oracle put its new tool, GraphPipe, into open source to standardize machine learning. (Pixabay)

Oracle released a new tool, which is called GraphPipe, into open source in order to speed up real-world deployments of machine learning.

GraphPipe, which Oracle has put into open source via GitHub, was designed to standardize and clarify machine learning models in order to scale out services and applications to customers.

Vish Abrams, an Oracle architect for cloud development, and his deep learning team created GraphPipe, which is comprised of a set of libraries and tools. In a company blog post, Abrams, who previously served as chief technology officer of OpenStack startup Nebula and as  principal engineer at Rackspace, said that there were three primary challenges that were stifling machine learning deployments.

FREE DAILY NEWSLETTER

Like this story? Subscribe to FierceTelecom!

The Telecom industry is an ever-changing world where big ideas come along daily. Our subscribers rely on FierceTelecom as their must-read source for the latest news, analysis and data on the intersection of telecom and media. Sign up today to get telecom news and updates delivered to your inbox and read on the go.

RELATED: LF Deep Learning signs up 5 more members, names AT&T's Gilbert as governing board chair

The first challenge is there's no standard model for serving APIs, "so you are likely stuck with whatever your framework gives you," according to Abrams.

"Your business application will generally need a bespoke client just to talk to your deployed model," Abrams said. "And it's even worse if you are using multiple frameworks. If you want to create ensembles of models from multiple frameworks, you'll have to write custom code to combine them."

The second challenge is building server models can be extremely complicated. Out-of-the-box solutions are few and far between, and deployments get much less attention than training, according to Abrams.

"Try building a GPU version of TensorFlow-serving, for example," Abrams said. "You better be prepared to bang your head against it for a few days."

The last challenge is many of the existing frameworks for machine learning don't focus on performance, which means they can fall short in certain uses cases.

"We created GraphPipe to solve these three challenges," Abrams said. "It provides a standard, high-performance protocol for transmitting tensor data over the network, along with simple implementations of clients and servers that make deploying and querying machine learning models from any framework a breeze. GraphPipe's efficient servers can serve models built in TensorFlow, PyTorch, mxnet, CNTK, or caffe2."

When paired with artificial intelligence and analytics, machine learning holds promise for internet of things applications, virtual reality, autonomous vehicles and the rollout of 5G services.

Telcos such as BT and AT&T are using machine learning and artificial intelligence to enable automation. In a recent FierceTelecom Q&A, Mazin Gilbert, vice president of advanced technology and systems at AT&T Labs, said machine learning, artificial intelligence and automation will lead to the creation of autonomous networks.

Suggested Articles

Expect a full dose of Kubernetes at next week's VMworld conference in San Francisco, according to VMware CEO Pat Gelsinger.

If there is indeed a global recession in the offing, no one has told the top cloud and colocation providers.

On the same day as its second quarter earnings, VMware announced it's buying Carbon Black and Pivotal for a combined value of $4.8 billion.