DeciA startup with 50 employees developing a platform to build and optimize AI-powered systems is led by Insight Partners today with the participation of Square Peg, Emerge, Jibe Ventures and Fort Ross. Announced that it has completed a $ 25 million Series B funding round. Venture company and ICON to raise the total amount of the company to 55.1 million dollars.Funds will be used for expansion DeciAccording to co-founder and CEO Yonatan Geifman, market development activities not only support the company’s R & D activities, but also.
Companies face some hurdles in creating AI models that analyze text, audio, and images to deploy across apps and services.The cost is very high — training a single model on commercial hardware can be costly Tens of thousands Of the dollar, if not more. While the new generation of chips and custom-designed AI accelerators have helped to ease some of the burden, building a model from scratch is still not an easy task.
Geifman proposes Neural Architecture Search (NAS) as a solution. NAS, a family of technologies that Deci relies heavily on, helps us automatically find the best model at a low cost for a particular problem.Deci isn’t unique with this — Google’s Vertex AI The service leverages the NAS to optimize model performance for specific customer-specified tasks. However, Geifman claims that the Deci platform provides low-cost access to NAS features.
In 2019, Geifman co-founded Deci with Ran El-Yaniv and entrepreneur Jonathan Elial. Geifman and El-Yaniv met in Technion’s Computer Science Division. Geifman was a PhD candidate and El-Yaniv was a professor.
“DeciUnique technology [can generate] A new image classification model that offers more than twice the execution time and accuracy compared to the most powerful models available, “Geifman emailed TechCrunch. “This means that we can deploy AI applications to the CPU that were previously only deployable on large and expensive GPUs.”
Those are lofty claims. However, with the help of Intel, Deci announced in March last year a strategic business and technology collaboration with startups to optimize machine learning on Intel processors. This partnership has created a model that accelerates the performance of question-and-answer tasks on Intel CPUs and an image classification model for the Cascade Lake processor that “significantly reduces computing overhead.”
Geifman previously told TechCrunch that one of Deci’s customers, a video conferencing provider, used the platform to deploy the ability to blur the background of a user’s device. Others theoretically have a GPU, and even if they have the computing power to do something, Deci to build a better model for their own internal computing. I am using it.
“”Deci It was created to empower developers and eliminate production-related bottlenecks throughout the AI lifecycle, “Geifman said. “The business impact of this feature is … faster to production, unlocking new AI use cases, and adapting to new market segments of resource-constrained devices.”
Geifman also states that compressed models help companies save inference computational costs, that is, the cost of actually serving the model after it has been deployed. Due to the popularity of cloud hosting models, more than one-third of businesses are up to 40% over their cloud budget. vote According to observable software vendor Pepperdata.
Geifman claims that Deci’s business continues to grow, but startups face challenges such as NAS technical limitations. (NAS, this is Difficult to evaluateIn addition, Deci is competing with many companies developing ways to make their models more efficient, such as OctoML, Neural Magic, and OmniML.
The next few months will be a test of Deci’s robustness against headwinds.
“We can’t disclose the valuation, but it’s a significant increase over the previous time. Due to Deci’s business growth and the opportunity to expand its product to additional domains such as natural language processing, it’s an existing one. Investors have decided to double to support their growth, “Geifman said. “I didn’t see a big impact. [from recent economic developments].. Our focus was primarily on enterprises, but the slowdown mainly affected middle-market and start-ups. “
Lonne Jaffe, managing director of Insight Partners, a board member of Deci, said in a TechCrunch email: A guide to finding alternative models that produce similar prediction accuracy with significantly improved efficiency … [It’s a value add because] hBy making the infrastructure of AI systems more efficient, AI products can be qualitatively different and better. It’s not just cheaper and faster to run. ”