Algorithmia aims to democratize deep learning with expansion of its marketplace
Source: Dan Richman

An application using deep learning applied color to this monochrome image based on what it “knows” of how colors appear in the world. (Via Algorithmia)

Deep learning — attempts to make software mimic the functions of the human brain — is one of the most conceptually difficult areas of computing, requiring programmers with highly specialized training. But Seattle-based startup Algorithmia is aiming to make deep-learning algorithms from the foremost researchers accessible to anyone wanting to harness the technology.

Algorithmia CEO Diego Oppenheimer

Algorithmia today is adding 15 deep-learning algorithms to its marketplace of roughly 2,000 callable APIs of all kinds, said Diego Oppenheimer, Algorithmia’s founder and CEO. Algorithms are sets of rules or procedures used to solve a computing problem. Companies building apps might use the deep-learning algorithms to identify faces in a crowd, detect nudity or gore in images, colorize images, or recognize stress in a human voice.

The company has also added access to the specialized computing needed to run the sophisticated algorithms.

“Our mission is to make the state of the art of machine learning discoverable by everyone,” he said. “For 99.9 percent of developers and companies, it’s been out of reach. We are the first company to provide these algorithms to any company that wants them.”

The new offerings benefit machine-learning researchers, too, Oppenheimer said. “It’s the first time university developers have a platform for doing massive distribution of their work,” he said.

Those wanting to use an algorithm simply call its API. The algorithm runs on one of the cloud services Algorithmia uses — Microsoft Azure, Amazon Web Services or Google Cloud Platform — and returns the results to the calling application. Each of those three services offers the unusual and complex mixture of CPU and GPU (graphics processing unit) computing required to run the matrix-based logic that characterizes machine-learning algorithms, Oppenheimer said.

If an algorithm is open source, Algorithmia charges only for the compute. It takes 30 percent of the royalty fee that the other algorithms’ authors charge.

Algorithmia raised $2.4 million in a round of seed funding led by Madrona Venture Group in 2014.