OctoML has announced the successful completion of its Series A funding round. The startup according to reports raised $15 million to help support its growth. The funding round was led by Amplify alongside Madrona Venture Group which led its $3.9 million seed round last October.
The platform was also backed by the team behind the Apache TVM machine learning project. The startup alongside TVM looks to facilitate the use of machine learning to optimize machine learning models. This, in turn, will enable them to run efficiently on various types of hardware configurations.
CEO of OctoML and professor at the University of Washington, Luiz Ceze disclosed that the project has achieved great strides in recent times, despite several challenges it had to surmount. One such challenges were finding out how to make use of it in the edge and clouds. The solution came in the form of the TVM project, now Apache incubating project. The incubating project started as research at Paul G. Allen School of Computer Science and Engineering a few years ago. And was later launched by Ceze alongside his research team at the university.
According to Ceze, TVM is a modern operating system for machine learning models. Adding that, “a machine learning model is not a code, it doesn’t have instructions, it has numbers that describe its statistical modeling”.
The project has gained the attention of several users with its framework already in use in huge tech companies like Amazon, ARM, AWS, Facebook, Intel, Google, Microsoft, Xilinx, and Nvidia. With OctoML tech, these companies seek to optimize their deep learning models for IoT/edge on several platforms like phones, health devices, cars, etc.
Despite the use of AI, human intuition is still required to pick the right model for specific hardware. Hence, the need for OctoML and its Octomizer Saas product. The Octomizer is a paid cloud-based solution.
The Octomizer basically allows users to upload their model to the service. While it automatically optimizes, benchmark and packages the models compatible with specified hardware and in its desired format. When optimized, the models operate significantly faster as they can now fully leverage the hardware they operate on.
Another benefit of the Octomizer is that more efficient models cost less to run in the cloud. They make use of cheaper hardware with less performance but still get the same results. According to reports, TVM currently already results in 80X performance gains in some cases.
Ceze pointed out that, “the secret sauce in our technology is to use ML to optimize ML, reducing the optimization and tuning time by orders of magnitude”.
At the moment the OctoML team comprises 20 engineers, with plans to add more competent engineers in the near future. As of now, Amazon Alexa’s wake word detection is also powered by TVM.
Conclusively, Ceze stated that “This is a hard time to be positive about the future but we see very good things ahead”.