By Erik Ottem-Cachengo, Innovative Technology Marketing Professional

Picking an Edge AI Application Development Platform doesn’t seem that complicated. Picking the wrong platform could impact your productivity and the transition from development and training to inference and production. As momentum continues to move to executing analytics at the data source, it is important to point out that the environment on the edge impacts platform technology choices. Let’s take a look at some of the considerations.

1. Decide your data strategy- where does the data come from – that’s where you want your Edge AI. Some data you might want to keep forever. Some data will be of transitory value. If you are subject to audits to justify your training models you’ll want to keep the data that was used in training available. Having a well thought out data strategy can help you avoid costly mistakes and wasted money.
   
2. Build the app on the same platform you’ll deploy the solution on. Performance and model behavior are important as you roll out your solution. The training platform should be similar to the production platform so you can appropriately judge how the model will work when moved into production.
3. Build the app on a platform that supports the entire application stack (e.g. custom databases, web servers, and caching). You need to support training and inference but don’t forget the other aspects of your application stack that also need support. Having a highly capable node designed for the edge is important for flexibility and ROI.
4. Edge environments need low power but lots of capability. This could be one of the biggest reasons that you want the development and production platform to be the same. The edge is a different world from the data center in terms of available power and space. Your platform choice has to reflect these realities of analytics at the edge.
5. Security is important, how do you get a secure network to protect your IP? Traditional approaches to security like a VPN can still leave you exposed to a man in the middle attack or spoofing. A zero trust network that authenticates end points is more secure and should be considered.

Picking an Edge AI platform does require balancing a number of factors, and hopefully this look into some of the considerations is helpful. Generic computing platforms for Edge AI may work, but it is not an optimized solution and can impact your time to solution and ROI. Commodity off the shelf (COTS) hardware makes sense in some environments but perhaps not at the edge. You want an edge system for your edge analytics. In today’s competitive environment you need every advantage, so it’s worth some time to explore your options.

 
Bio: Erik Ottem-Cachengo has over 30 years in technology sales, marketing, and product management, and has held positions in large and small companies in Silicon Valley. He was early into storage area networks with Gadzoox and Agilent. Early into all flash arrays with Violin Memory. Early in Object Storage with HGST. Now he sees the opportunities in edge analytics with Cachengo. By bringing intelligence to the edge, real time analytics are easy and can lead to improvements in operations, manufacturing, medical, agriculture, transportation and many other industries. Erik holds an MBA from Washington University in St. Louis and a BS from the University of California, Davis.

Related:

• What Makes AI Trustworthy?
• Budgeting For Your AI Training Data: Consider These 3 Factors
• A Critical Comparison of Machine Learning Platforms in an Evolving Market