Recurrent Dynamics, a Toronto based AI start-up announced today the implementation of world’s first platform based on scatter training, a faster way to develop AI services at no upfront data collection and training costs.
The Skatr training platform utilizes a newly developed scatter training method. According to Recurrent Dynamics CEO and founder Max Trokhimtchouk “cloud training or federated learning utilizes batch training process, in contrast, scatter training is a continuous process on devices at the edge, which allows for faster model prototyping.”
“5 weeks ago, we announced a computational breakthrough that enabled AI to be trained on small cheap devices such as cell phones and tablets, rather than expensive cloud infrastructure. Today we have moved this technology from theory to practice by implementing a fully functional working prototype on distributed Raspberry Pi devices, each training AI continuously and sharing their learning,” stated Mr. Trokhimtchouk.
The Skatr platform’s advantage of faster model prototyping with no upfront data collection and training cost, enables small teams to be more efficient than current leaders utilizing cloud based training infrastructure. As a result, the number of AI applications and services utilizing the Skatr platform will explode. This will put competitive pressure on existing cloud infrastructure providers and VC funded AI start-ups utilizing cloud training.
Cloud infrastructure providers to lose revenue as AI computational workload shifts from central servers to edge devices.
It may seem inconceivable today, but the multi-billion dollar cloud infrastructure supporting the AI ecosystem is at risk of disruption. It is a blind side for providers such as AWS, Google Compute Cloud, and Microsoft Azure. Training of AI is going to move to the edge. Recurrent Dynamics Skatr training platform provides a leading first step in that move. According to a recent Bloomberg article, this could create a 4 trillion dollar market.
Valuation for AI first movers may drop to zero, if they fail to adapt.
According to CB insights, AI 3,600+ start-ups have received $66 Billion dollars in funding since 2013. One of the reasons for these large investments is the high upfront cost of cloud training, which includes data collection, and computational infrastructure. The Skatr platform will remove this barrier to entry by eliminating upfront cost.
Mr. Trokhimtchouk, who has 12 years experience in machine learning, and a PhD in mathematics from Berkeley stated, “It is with great excitement that we make this announcement today. Skatr is a new peer-to-peer continuous training platform that will reshape the AI landscape.”
About Recurrent Dynamics
Recurrent Dynamics is a Toronto, Canada based AI startup. Our mission is to advance the capabilities of Artificial Intelligence. Go to www.RecurrentDynamics.com to learn more and register to receive updates.