UTMIST is proud to present the Entrepreneurship in AI Speaker Series, where we invite founders and researchers from exciting AI companies to share with us their technologies and startup journeys.
Our first guests are Untether AI, a tech startup from Toronto that rethinks how computation for machine learning is accomplished and creates ultra-efficient, high-performance AI accelerators that enable new frontiers in AI applications. Founded in 2018, Untether AI has raised over $27M and is backed by Radical and Intel.
Watch the event recording to learn about the story behind Untether AI and the technology behind the most efficient AI compute machines on the market. You will also be exposed to the mathematics and physics behind neural networks and quantization.
Dr. Martin Snelgrove, CTO at Untether AI, has over 20 years of executive level leadership experience, and also served as a professor and researcher at the University of Toronto and Carleton University for 16 years. An expert in bringing academic R&D to commercialization, he is widely acknowledged as a pioneer in the field of digital, signal processing, and mixed signal processing. Dr. Martin Snelgrove is an author of 36 patents and over 100 refereed publications.
Darrick Wiebe, Head of Technical Marketing, has developed softwares in a wide range of markets including military and data centres. He previously founded two successful startups developing and leveraging graph database application framework technology.
Dr. Paul Grouchy, Senior Deep Learning Application Engineer, has significant R&D experience in both academia and industry, applying cutting-edge AI techniques to myriad problem domains, including terrestrial and space robotics.
AI workloads require increasing amounts of compute resources. The slowing of Moore’s Law and the end of Dennard scaling is limiting the potential from traditional approaches of computing. Untether AI was founded to radically rethink how computation for machine learning is accomplished. In current architectures, 90 percent of the energy for AI workloads is consumed by data movement, transferring the weights and activations between external memory, on-chip caches, and finally to the computing element itself. By focusing on the needs for inference acceleration and maximizing power efficiency, Untether AI is able to achieve at-memory computation and deliver two PetaOperations per second (POPs) in a standard PCI-Express card form factor.
Interested to learn about the theory and implementation behind AI computation and at-memory computation? Come and meet the people behind the most efficient AI accelerator on the market, to learn about how they accomplish this feat.