Moving AI from Cloud to On-Premise

I grew up in a technological world where on-premise software – namely, software installed on a company’s own machines and behind a firewall – was the only available option. In the past decade, cloud computing has grown in popularity with the promise of saving time and money while offering greater flexibility, scalability, and convenience to the user.

AI is no exception to this trend. With Microsoft, Amazon, and Google, among others, offering easy to use cloud-based AI developing tools, complete with pre-trained Deep Learning and AI models,  cloud-hosted deployment options are simple to deploy.

While this is great for the Amazons, Googles, and Microsofts of this world, it is not so great for their users. Convenience comes at a price: cloud computing can be expensive. But more importantly, issues such as portability of your solution (what happens if you don’t want to use Google anymore?...), security, latency, and the fact that cloud-based AI does not bode well with many modern and real-world uses of AI.

More often than not, AI is applied to a compute Edge: from a smart phone whose camera and mics are AI-augmented, to a drone that needs to steer away from obstacles and interpret its camera input, to robots navigating grocery stores and logistic facilities, all the way to cameras inside large production centers and industrial machines .In all these cases, a vast amount of data is generated at a compute Edge that needs quick, real-time interpretation.   

Cloud AI is not the right fit for:

  • the staggering and unrealistic connectivity requirements (what happens if your Wi-Fi and/or 5G is out for 2 seconds and your cameras goes dark!)
  • latency(you can’t expect 20ms latency from image acquisition to AI requirements to be satisfied by a cloud provider)
  • pay-per-image economics (imagine how quickly the costs add up)
    Edge AI is required in these real-world scenarios.
Additionally, many manufacturers do not want to part with their proprietary ‘unfiltered’ data, , and give them to a cloud provider. They want control over the data, raw images of their products and facilities, and the hardware implementation. Edge AI natively fulfils these needs.

Trends come and go. While AI was quick to migrate to the cloud, in reality its users want it on site, close to them, and in the privacy of their premise.

Read more about the new Neurala VIA software for on-premise visual quality inspection. 

My parents always told me “do not keep your head in the cloud all the time”. They were right.

Watch this space for real life on-premise use cases in our new blog series: Customer Insights