A machine learning engineer needs to select a deployment strategy for a new machine learning application. The feature values are not available until the time of delivery, and results are needed…

Question

Options

• AEdge/on-device
• BStreaming
• CNone of these strategies will meet the requirements.
• DBatch
• EReal-time

• A
  6% (1)
• B
  11% (2)
• E
  83% (15)

Explanation

Real-time (online) inference is the correct strategy because it processes individual records on-demand at the moment of prediction, which satisfies both constraints: features are computed at delivery time (not pre-computed), and latency is minimized for single-record responses.

Why the distractors fail:

• Batch (D) processes many records together on a schedule - it requires features to be available in advance and doesn't deliver "exceedingly fast" results for a single record.
• Streaming (B) handles continuous data flows in near-real-time but is optimized for pipelines of events, not instant single-record lookups; it also typically introduces pipeline latency.
• Edge/on-device (A) is a deployment location, not a deployment strategy in the same sense - it describes where the model runs, not how inference is triggered. It could support real-time inference, but alone it doesn't answer the strategy question.
• C is wrong because real-time satisfies both requirements.

Memory tip: Think of the two constraints as a filter - "one record at a time" eliminates Batch, and "features not available until delivery" eliminates any pre-computation approach. What's left standing is Real-time inference, which waits for the live request before doing any work.

No community discussion yet for this question.