CoCoPie, which optimizes AI for edge devices, raises $6M

CoCoPie, a startup developing a platform to optimize AI models for edge devices, today announced that it raised $6 million in a series A funding round led by Sequoia China Seed Fund. The capital, which values the company at $50 million post-money, will be put toward R&D and growing CoCoPie’s customer base, according to CEO Yanzhi Wang.

The market for edge AI — a combination of AI and edge computing, which captures, stores, processes, and analyzes data near the source — is expected to be worth $1.83 billion by 2026. A Markets and Markets report attributed the growth to various factors, but particularly the increased adoption to remote working capabilities, remote asset maintenance and monitoring, factory automation, and telehealth. Research from Google Cloud and The Harris Poll found that the pandemic led to “significant” increase in AI adoption across manufacturing supply chain, inventory, and risk management, for example.

CoCoPie, which was founded in 2020, offers a product that enables AI models to run on off-the-shelf devices through techniques including model compression. Through “pattern-based” pruning and “pattern-aware” code generation, the platform generates efficient execution codes, leveraging a framework to identify which parts of models to prune without impacting accuracy.

“The pandemic has highlighted why a software solution like CoCoPie is such a necessity. There are tens of billions of existing devices powered by older generation processors, most of them not able to run AI models without CoCoPie’s technologies,” Wang, who cofounded CoCoPie with fellow academics Xipeng Shen and Bin Ren, told VentureBeat via email. “Facing the global chip shortage, many of these processors in existing devices can be recycled; with CoCoPie’s solution on top, these older generation chips can even outperform newer generation processors.”

Optimizing models

AI optimization tools aren’t particularly novel. Neural Magic and OctoML — which Wang considers competitors — offer solutions comparable to CoCoPie’s platform. But Wang said that CoCoPie’s technology can uniquely enable speedups on existing mobile device as well as low-end internet of things processors, microcontrollers, and digital signal processors.

Benchmarked against libraries like PyTorch Mobile and TensorFlow Lite, CoCoPie claimed that it can achieve inference speeds of between 6.7 milliseconds and 11.8 milliseconds on a Samsung Galaxy S10 smartphone (with a Qualcomm Kyro 485 GPU and Adreno 640 GPU) with around 78% accuracy. Moreover, CoCoPie said it can make computer vision inference as fast as 3.9 milliseconds — equivalent to 256 photos per second — on a mobile device, or up to 331% faster than PyTorch Mobile.

“Overall, CoCoPie technology can improve the end-user’s mobile experience. From faster performance speeds to improved zoom on mobile devices, CoCoPie’s technology is going to revolutionize the smartphone industry,” Wang said.

Sid Nag, research VP at Gartner, said that the growth of startups like CoCoPie signals that the next frontier in IT services will be driven by the nexus of cloud, edge, 5G, AI, internet of things, and data and analytics. This nexus, he said, will enable solutions for use cases like smart cities and retail, drones, connected and autonomous vehicles, and remote health care.

“Use of AI in edge computing for a specific use case such as smart retail is an example of that nexus – by leveraging edge computing solutions that are located in store and retail outlets, where infrastructure needs such as compute is available locally to these edges, allows retail applications such as point of sale and inventory control to run closer to end users,” Nag told VentureBeat via email. “In addition, technologies such as AI can aid in pattern recognition of customer-buying behavior to drive additional upsell opportunities. AI can also be used to identify patterns of usage to aid in strategic placement of edge nodes.”

Fifteen-employee CoCoPie has more than 10 customers, including Tencent and Cognizant, and was recently awarded a National Science Foundation Small Business Innovation Research grant for $250,000. CoCoPie said it will use the grant to “further enhance” its core optimization technologies.