OctoML Inc., a fresh-faced machine learning startup recently spun off from the University of Washington, today announced that it has raised $3.9 million in funding to tackle the complexity of deploying artificial intelligence software.
Setting up an AI model on a hardware system is much different than the typical application install. To maximize an algorithm’s performance and power-efficiency, engineers must painstakingly optimize their code for the specific chip powering the host system. That’s not always a feasible option for development teams, particularly at companies with multiple neural networks running on several different types of hardware.
OctoML is looking to make the task less resource-intensive. The startup’s 10-strong team, led by Chief Executive Officer and University of Washington professor Luis Ceze (second from the left), has developed an open-source toolkit called Apache TVM that can automate the model deployment process. It uses machine learning to optimize neural networks according to the constraints of each platform on which they’re installed.
TVM supports a wide range of hardware environments. The software enables engineers to deploy models on most of everything from smartphones to the specialized, AI-optimized accelerator chips companies increasingly use in their data centers to support machine learning workloads.
“Apache TVM’s machine-learning based approach to optimizing machine learning systems fundamentally enables targeting a constantly changing and expanding set of hardware targets such as data centers, cars, phones, health devices and embedded systems with much less engineering,” said OctoML Chief Technology Officer Tianqi Chen (second from the right.)
TVM has gained a lot of traction in the machine learning ecosystem. Major tech firms such as Amazon.com Inc., Microsoft Corp. and Facebook Inc. use the software internally to tune-up their AI models, while chipmakers including Qualcomm Inc. are contributing to the project’s source-code.
OctoML plans to monetize TVM’s popularity with an upcoming managed cloud service that uses the technology to ease AI developers’ work. The offering will “lower engineering and operating costs for customers, and lowers the risk of dependence on specific platforms,” promised CEO Louis Ceze without going into detail.
The startup’s newly closed $3.9 million round was led by early-stage investor Madrona Venture Group with participation from Amplify Partners. OctoML will use the capital to hire more engineers and lay the go-to-market foundations for the launch of its commercial offering. Photo: OctoML