Deci, a deep learning platform company that helps enterprises build AI models to run on different types of hardware and service environments, has $21 million Series A funding.
As more companies want to work with AI, there remains a gap between lab-based developments and getting AI into a production environment on the right hardware. Deci CEO Yonatan Geifman told Fierce Electronics some deep learning platforms use “pruning techniques” to optimize their ability to run on the AI chip hardware currently available in the market. Deci takes a different route.
As Geifman explained, the company’s solution is based on “Neural Architecture Search (NAS) that incorporates both quantization and compilation technologies. This is a major differentiator from other deep learning acceleration solutions on the market, which utilize pruning techniques for optimization.”
He added that those pruning techniques do not directly accelerate sparse matrix multiplication, and are limited in their acceleration potential. Deci can achieve better results with its hardware-agnostic approach, he said.
“Deci’s solution represents a paradigm shift as it enables AI developers to seamlessly build deep learning models that address all production requirements from the get go,” Geifman said. “Models generated by Deci’s proprietary AI engine (AutoNAC) have been proven to deliver superior accuracy and performance compared to any other known state of the art models in the market today. Leveraging the same process, Deci optimizes existing models as well. Its platform includes a complementary set of tools that help to streamline additional stages in the development lifecycle.”
That translates to an ability to move to a production environment with AI models in a matter of days without spending additional time manually designing models. That time can now be spent working on their specific use cases.
Deci also claims to boost AI inference performance, enabling new use cases and helping reduce compute costs by up to 80%. The funding announcement follows a collaboration between Deci and Intel at an MLPerf benchmark event in late 2020, where Deci’s AutoNAC accelerated the inference speed of the ResNet-50 neural network on several Intel CPUs. That led to a broader strategic collaboration between the two companies earlier this year with the aim to optimize deep learning inference on Intel Architecture CPUs.
“Besides Intel, we are currently working with multiple enterprises across industries, including global semiconductor manufacturers, industry-leading video conference solutions, smart city developers, and AI-based medical diagnostic applications,” Geifman said.
He added that Deci’s AutoNAC technology has also been tested and approved by Hewlett Packard Enterprise, and that Deci is a member of the AWS Partner Path program.
The funding round was led by global private equity and venture capital firm Insight Partners. Deci’s existing investors Square Peg, Emerge and Jibe Ventures participated in the round and were joined by new investors including Samsung Next, Vintage Investment Partners, and Fort Ross Ventures. The investment comes 12 months after Deci secured more than $9 million in seed funding and brings the total funding to $30.1 million.
RELATED: Nvidia projects are helping AI find its human-like voice