Home › GCP ML Engineer › cloudcomputing › mleserving › Which Vertex AI Feature Store capability should …
Which Vertex AI Feature Store capability should an ML Engineer use to FETCH a feature vector for an incoming request at SERVE TIME — guaranteeing consistency with the features used during training?
ACloud DNS
BVertex AI Feature Store online serving (with parity to the offline store used in training)
CCloud Memorystore
DRecomputing features on the fly differently from training code
Answer & Solution
Correct answer: B. Vertex AI Feature Store online serving (with parity to the offline store used in training)
Feature Store online serving is the train/serve parity primitive (per PMLE §4.2). Recomputation creates training-serving skew; the others aren't feature stores.
Related questions
Which GCP service should an ML Engineer use to serve a model directly inside BigQuery as aWhich Vertex AI capability should an ML Engineer use to serve XGBoost / LightGBM / scikit-Which GCP-compatible streaming runtime should an ML Engineer use to MAKE PREDICTIONS over Which Vertex AI feature should an ML Engineer use to deploy a CUSTOM container that wraps Which Vertex AI deployment option should an ML Engineer choose for a HIGH-THROUGHPUT onlinWhich ML model-optimisation technique should an ML Engineer use to REDUCE inference latencWhich Vertex AI endpoint type should an ML Engineer recommend to keep model traffic INSIDEWhich Vertex AI feature should an ML Engineer use to run an A/B test between two model ver