Question

Your ML team serves a real-time fraud detection model on Vertex AI. Multiple teams independently compute customer features (e.g., 30-day transaction totals) in their own pipelines, and you've discovered that the feature values used at training time differ from those fetched at prediction time, causing training-serving skew. The team also wants to reuse these features across three other models. Which approach best addresses both the skew and the reuse requirements?

Accepted Answer

Adopt Vertex AI Feature Store: ingest features once, use point-in-time lookups for training data, and serve the same features via the online store at prediction time.

Answer

Store precomputed features in a shared BigQuery table and have each model query it directly during both training and online prediction.

Answer

Have each team persist its features to Cloud Storage as CSV files and load them into the model container at startup.

Answer

Recompute all features inside the serving container using the same transformation code to guarantee consistency.