Scaled Inference Wants to Bring Machine Learning to the Masses

 Artificial Intelligence  

Scaled Inference, a new company started by two ex-Google employees, is building machine learning technology similar to what’s used internally by companies like Google, making it available as a cloud service that can be used by anyone.

According to their website, Scaled Inference, is "enabling a new generation of intelligent software built by the masses and powered by an open shared platform."

The eight month-old company started by two ex-Google employees, is developing machine learning software as a cloud service.  They are automating the process so that the machines themselves will choose their method of solving a particular problem rather than relying on human experts to do it for them.

"We want to remove experts from the loop almost completely, where machines are given the data and the problem–the problem of detecting, say, [whether an image is] cats–and the machine can itself figure out what [mathematical] model would work best."

The result will be a system that can compute a variety of problems involving artificial intelligence with nearly any kind of data.

Scaled Inference has yet to launch a public product, but it expects to have a working prototype together soon and is now bringing together companies for a closed trial of the platform.

Olcan Sercinoglu–the co-founder of Scaled Inference Inc., told the WSJ he thinks biological systems are fundamentally different from machines and can’t be directly compared.

Scaled Inference has now raised $13.6 million at a valuation of about $60 million to build the software.

Scaled Inference is approaching the problem differently than competitors, according to Sercinoglu, by using a mathematical approach that takes advantage of machines’ ability to calculate rather than trying to emulate biological systems, about which he said too little is known.

“We want to remove experts from the loop almost completely,” he said, “where machines are given the data and the problem–the problem of detecting, say, [whether an image is] cats–and the machine can itself figure out what [mathematical] model would work best.”

Related articles Stanford Researchers and Google Create World's Largest Artificial Neural Network  The Clever World of Artificial Intelligence  Stuart Armstrong and Nick Hawes Discuss the Future of Artificial Intelligence

Scaled Inference’s software is said to be able to solve multiple types of problems with credit card data, using anomaly detection to find fraud and pattern detection to predict events such as what a customer is likely to purchase, which is unusual for a single system, he said.

“There are not enough people who can do it well, and companies like Scaled Inference are in a position to reduce the level of expertise needed to do machine learning,” said investor Vinod Khosla, adding that “machine learning systems introduce the need for machine learning science,” which is what Scaled Inference is working on.

“Ultimately, the ability of machines to solve more general problems, general artificial intelligence in the future, is really through getting down at principals and solving the model selection problem,” he said.

Sercinoglu and his co-founder, Dmitry Lepikhin, both spent several years working at Google where Sercinoglu started as an intern to Google Senior Fellow Jeffrey Dean and others.

He joined Google full time in 2002 and was an early contributor to Google’s infrastructure, which allows developers to routinely use tens of thousands of machines. He also worked on Google Borg, which manages Google’s servers; Google’s search infrastructure, which indexes Web pages; and Google Trends, which computes what people are searching for.

Then he moved into Google Research, where he reported to Peter Norvig, an expert in artificial intelligence, and became interested in how to scale artificial intelligence and machine learning across thousands of machines.

He left Google to start Scaled Inference in May.


SOURCE  Wall Street Journal

By 33rd Square

Embed