The company, whose technologies are built on SRI International's machine learning (ML) methodology research on how to deploy computer-based systems in complex situations, announced the Latent AI Efficient Inference Platform (LEIP) and its first platform product, LEIP Compress, a quantization optimizer for edge devices that enables smart and efficient IoT applications. LEIP is designed to enable developers working on edge AI projects to optimize within compute, energy, and memory budget without requiring changes to existing AI/ML infrastructure, processors, or development frameworks.
"One of the major goals for Latent AI is to simplify and accelerate the AI workflow from training to deployment," says Jags Kandasamy, CEO and Co-Founder of Latent AI. "With many thousands of compelling edge AI applications yet to be developed, we are helping make the development process as efficient and seamless as possible."
The company's underlying technology was initiated under DARPA projects and, says the company, represents a fundamental shift in the way AI is trained and computing resources are dynamically determined for highly resource-constrained hardware. It can compress large neural networks up to 90% with less than 1% accuracy loss.
In addition to the platform launch, the company announced it had recently closed a $3.5 million seed round of financing, led by Steve Jurvetson at Future Ventures and including industry leaders such as SRI International Ventures, Perot Jain. and Gravity Ranch.