How to Resolve Persistent Training-Serving Skew Alerts After Retraining?

You received a training-serving skew alert from a Vertex AI Model Monitoring job running in production. You retrained the model with more recent training data, and deployed it back to the Vertex AI endpoint, but you are still receiving the same alert. What should you do?

Update the model monitoring job to use a lower sampling rate.
Update the model monitoring job to use the more recent training data that was used to retrain the model. Source Reference Answer
Temporarily disable the alert. Enable the alert again after a sufficient amount of new production traffic has passed through the Vertex AI endpoint.
Temporarily disable the alert until the model can be retrained again on newer training data. Retrain the model again after a sufficient amount of new production traffic has passed through the Vertex AI endpoint.

Community Votes

78%

22%

78% of anonymous learners picked answer B. Votes are pick records left by other test-takers — they are not the verified answer.

Community Insight

Tests whether you know that model retraining does not automatically update the monitoring baseline, trapping candidates into retraining loops instead of updating the monitoring configuration.

This question tests understanding of how Vertex AI Model Monitoring calculates baselines for training-serving skew detection. The community consensus confirms that updating the monitoring job’s training dataset reference is required to align the baseline with newly retrained models and eliminate false positives.

Option D is the most frequent distractor because candidates assume waiting for new traffic and retraining again will resolve the skew, overlooking that the monitoring baseline must be explicitly updated to match the current training dataset.

Community Discussion (10 comments)

8619d79 👍 1 Selected: D

This approach acknowledges that the skew might be due to a mismatch between the training data and the current production data distribution. By waiting for sufficient new production traffic, you can collect a more representative dataset that reflects the current state of the production environment. Retraining the model on this new data ensures that the model is better aligned with the production data distribution, which should resolve the skew.

lunalongo 👍 1 Selected: C

C) - The model is adapting to the changing data distribution in production - Disabling alerts temporarily gives model a chance to adjust to new data *A) would hide/mask the skew; B) doesn't make sense because the monitoring job already uses the most recently trained data, it's just different from production data; D) is reactive and short term solution

AzureDP900 👍 1

C. Temporarily disable the alert. Enable the alert again after a sufficient amount of new production traffic has passed through the Vertex AI endpoint. Here's why: You've already retrained the model with more recent training data and deployed it back to the Vertex AI endpoint, but the alert persists. This suggests that the model is still adapting to the changing data distribution in production. Temporarily disabling the alert will give the model a chance to adjust to the new data distribution before the monitoring job starts firing alerts again. Once enough new traffic has passed through, you can re-enable the alert and continue monitoring the model's performance.

info_appsatori 👍 2 Selected: B

The baseline is calculated when you create a Vertex AI Model Monitoring job, and is only recalculated if you update the training dataset for the job.

SahandJ 👍 2

Is B actually the correct answer? According to the documentation, training-serving skew detection can only be enabled if the original training data is available. Furthermore, the baseline is automatically recalculated when the training data is updated. So does this question imply that the model is trained on data without updating the original training-dataset? If so then B is clearly correct. If they updated the training dataset with new data and then retrained the model then the model monitoring job's baseline should automatically have been recalculated. I see no other valid answers in that case?

pinimichele01 👍 1 Selected: B

This option can help align the baseline distribution of the model monitoring job with the current distribution of the production data, and eliminate the false positive alerts.

36bdc1e 👍 3

B This option can help align the baseline distribution of the model monitoring job with the current distribution of the production data, and eliminate the false positive alerts.

BlehMaks 👍 3 Selected: B

the cause of the issue could be that the developer forgot to switch their monitoring job to the latest training dataset and the monitoring job still compares prod data with old training dataset and they of course have a skew

pikachu007 👍 1 Selected: B

B. Update the model monitoring job to use the more recent training data that was used to retrain the model: This option directly aligns the model monitoring with the recently retrained model and ensures that the monitoring job reflects the characteristics of the latest training data.

b1a8fae 👍 1 Selected: D

A. Changing the sampling rate affects not training skew but cost efficiency: https://cloud.google.com/vertex-ai/docs/model-monitoring/overview#considerations B. The model monitoring job is already using the most recent data to detect skew. C&D are the same, except for D being more specific, so I would tend towards D.

Comments & Corrections

No comments yet — spotted an error or have a note? Share it below.

Expert Analysis

Why the Answer Is Correct

Vertex AI Model Monitoring establishes its baseline distribution using the specific training dataset provided during job creation. When a model is retrained with newer data, the monitoring job continues comparing live traffic against the original baseline, triggering persistent skew alerts. Updating the monitoring job to reference the latest training dataset forces a baseline recalculation, aligning the expected distribution with the current model and resolving the false positive alert. As noted in comment [3], the baseline is only recalculated when the training dataset for the monitoring job itself is updated.

Why the Other Options Are Wrong

Option A incorrectly suggests adjusting the sampling rate, which only impacts cost and statistical confidence, not the underlying data distribution mismatch. Options C and D propose disabling alerts or waiting for more traffic, which ignores the root cause: the monitoring baseline is outdated relative to the newly trained model. Disabling alerts merely masks operational issues rather than fixing the configuration drift between the model and the monitoring job. Comment [10] correctly points out that sampling rates do not address training-serving skew detection logic.

Community Comment Notes

Candidates consistently highlight that developers often forget to sync the monitoring job’s training data reference after model updates. Comment [2] emphasizes that the monitoring job compares production data against the old training dataset, causing unavoidable skew until the reference is updated. Comment [4] validates this behavior by referencing official documentation stating that baseline recalculation occurs specifically when the monitoring job’s training dataset is modified. These insights reinforce that configuration alignment, not additional retraining cycles, is the correct remediation step.

Official Reference

Exam Strategy

Always verify whether a configuration change requires explicit updates across dependent services, especially in managed ML platforms. In certification exams, questions about monitoring and alerts often test baseline synchronization rather than iterative model retraining cycles.

Related Analysis

Practice All PMLE Questions

Access 65 questions with complete answers and detailed explanations.

View Full PMLE Practice Test →

← Back to PMLE Study Guide