Building and deploying artificial intelligence (AI) models at the network edge is a cumbersome process today. AI models must first be trained before they are released to power thousands (if not millions) of low-cost, low-power edge device hardware. And these models are often trained in the cloud or on other large-scale data center environments with static, pre-collected, pre-labeled data.

That's all well and good until the models are deployed in real-life operating environments where data and information are anything but static, pre-collected and pre-labeled! These models are suddenly expected to operate in these new environments and rapidly identify edge cases or anomalies. This causes chaos in the model and results in continued degradation of performance until the model becomes totally unusable.

New techniques and tools in AI deployment allow organizations to move their model development to a continuous learning cycle while drastically reducing operational costs and bandwidth requirements. The goal is to make it simpler for AI models to adapt to changing conditions at the edge - closer to the point where data is being created and consumed.