Apache TVM Creators Launch OctoML to Help Companies Use Deep Learning

Tatum Hunter
October 23, 2019
Deep learning model optimization compiler
photo via shutterstock

We humans are great at problem-solving, but we can’t do math fast enough to analyze large data sets.

Machine learning, a type of artificial intelligence, solves that problem. Machine learning occurs when a computer uses algorithms to process a large amount of data and search for patterns that may be helpful to its human overlords, like which markets are ripe for expansion or which new features would boost engagement with an app.

Sometimes, however, data is too complex for traditional machine learning to pull out any insights. The most often-used example is image recognition. Imagine you’ve collected millions of images of hooved animals and want to sort them by species. An algorithm can process a lot of image data quickly, but it’s not smart enough to tell the difference between a goat and a horse. Humans can tell the difference, but they can only look at one image at a time.

That’s where deep learning comes in. Deep learning lets computers process huge, unstructured data sets by passing them through layers of algorithms. The first layer will derive the most general insights: is there a figure in this image? The last layer will derive the most specific ones: does that figure have horns, a goatee and cloven hooves?

Deep learning models can be extremely helpful for researchers and corporate data scientists looking for solutions to complex problems. But these models are built for particular hardware frameworks, and adjusting them to fit what’s available requires time, expertise and money. 

What these parties need is a program that automatically optimizes deep learning models to work with diverse hardware setups. 

That program arrived with the advent of Apache TVM, an open source project that grew into an automated deep learning model optimization and compilation stack. Its originators are using the project’s advancements to launch OctoML, a service that will help companies deploy deep learning models for cloud and edge computing. 

OctoML announced today it raised a $3.9 million seed round led by Madrona Venture Group with participation from Amplify Partners. OctoML’s CEO Luiz Ceze is a partner at Madrona.

Huge tech companies such as Amazon, Facebook, Microsoft, Xilinx and Qualcomm already use Apache TVM to streamline their use of deep learning models. The launch of OctoML will allow its team to help companies of all sizes use Apache TVM to deploy and monitor deep learning systems without sky-high engineering and operating costs. 

The company, which spun out from the University of Washington Allen School for Computer Science and Engineering, currently has a team of 10 and plans to double its headcount in light of the new funding, a spokesperson told Built In. Its website lists open positions for a systems software engineer, full-stack engineer and Systems Modeling Language engineer.

Seattle startup guides

Best Companies to Work for in Seattle
Coolest Tech Offices in Seattle
Best Benefits at Seattle Tech Companies
Women in Seattle Tech