Part 10/17:
Addressing where to run models depends on data volume and latency requirements:
Edge Computing: Essential for high-volume, real-time tasks like suspicious activity detection across multiple cameras. Running models on local hardware minimizes latency and bandwidth issues.
Cloud Computing: Suitable for less frequent, large-volume batch analysis such as historical data review or inventory audits. Model training, validation, and updates often occur here.
The team proposes practical rules of thumb to guide infrastructure investments, considering factors like data volume, real-time needs, and operational costs.