A deployed model has access to its original training data, and the team wants to detect whether production feature distributions are deviating from that training distribution. What is this called in Vertex AI Model Monitoring v1?

Question

Accepted Answer

A. Training-serving skew detection. Training-serving skew detection compares production feature distributions to the feature distribution used for training. Grounding is a generative AI technique for tying responses to sources, so it does not describe this monitoring objective.

A deployed model has access to its original training data, and the team wants to detect whether production feature distributions are deviating from that training distribution. What is this called in Vertex AI Model Monitoring v1?

Related Questions

Discussion