Refer to the Exhibit Practice Questions

Remove dropout layers

Why wrong: Removing dropout would remove a regularization technique and likely increase overfitting.

Increase learning rate to 0.01

Why wrong: Increasing the learning rate may cause the model to converge too quickly or oscillate, often worsening overfitting.

Increase units in the first dense layer to 512

Why wrong: Increasing units increases model capacity, which typically worsens overfitting.

The auditor role lacks 'evaluate' permission

Why wrong: Auditor permissions are for viewing, not deployment.

Fairness check threshold is set to 0.8, which is too low

Why wrong: Threshold does not cause bypass, it only measures fairness.

External_user role can perform inference, which triggers unfair predictions

Why wrong: Inference does not bypass fairness check.

The intern's role is not included in the allowed roles

Why wrong: That alone wouldn't cause denial if not explicitly denied.

The policy JSON has a syntax error

Why wrong: The JSON is valid.

The endpoint path is incorrect

Why wrong: Exhibit shows correct path.

Remove all demographic features from the model.

Why wrong: Removing features may not eliminate bias if other correlated features exist.

Use a different algorithm that is inherently unbiased.

Why wrong: No algorithm is inherently unbiased; bias depends on data and deployment context.

Hire more diverse data scientists.

Why wrong: Diversity in team is beneficial but does not directly fix model bias.

A required feature is missing from the production data pipeline.

Why wrong: No mention of missing features; all features are present.

The model was overfitted to the training data.

Why wrong: Training and test accuracy are close (0.87 vs 0.85), so overfitting is unlikely.

The model's confidence threshold needs to be adjusted.

Why wrong: Adjusting threshold changes precision-recall trade-off but does not fix the underlying drift.

Increase n_estimators to 500.

Why wrong: More trees may still overfit if each tree is deep.

Set max_depth to None to allow trees to grow fully.

Why wrong: Increasing depth often increases overfitting.

Switch from RandomForest to a linear model like LogisticRegression.

Why wrong: Changing algorithm is not a parameter change and may underfit.

The precision metric is not included in the evaluation script

Why wrong: Precision is listed in metrics, so it is computed.

The deployment target is set to staging instead of production

Why wrong: The target is 'production', so that's not the issue.

The operator manually overrode the threshold

Why wrong: No evidence of manual intervention in the exhibit.

Principal component analysis (PCA)

Why wrong: PCA reduces dimensionality and is not applied to the target variable.

Standardization (Z-score)

Why wrong: Standardization centers and scales data but does not reduce skewness.

One-hot encoding

Why wrong: One-hot encoding is for categorical variables, not for transforming a continuous target.

The event will be stored in a dead-letter queue automatically.

Why wrong: A dead-letter queue requires explicit configuration, not automatic.

The event will be accepted but user_id will be set to null.

Why wrong: Required fields cannot be null.

The event will be accepted with a default user_id of 0.

Why wrong: No default value is defined in the schema.

Ensure equal error rates across groups

Why wrong: Incorrect; equal error rates refer to equality of opportunity, not demographic parity.

Ensure equal accuracy across groups

Why wrong: Incorrect; demographic parity focuses on prediction rates, not accuracy.

Ensure model interpretability

Why wrong: Incorrect; interpretability is about understanding model decisions, not fairness metric.

Model accuracy below threshold

Why wrong: Overall accuracy is 0.98, which is high and above typical thresholds.

High latency

Why wrong: The average latency of 45 ms is acceptable for real-time applications and not a critical concern.

Low throughput

Why wrong: Throughput of 1200 requests per second is high and not an issue.

Training hyperparameters

Why wrong: Hyperparameters are already included.

Model type

Why wrong: Model type is already specified.

Evaluation metrics

Why wrong: Metrics can be defined separately; preprocessing is more fundamental.

Review the fairness check settings to ensure protected attributes are still relevant.

Why wrong: Fairness is separate; drift detection triggered due to feature drift, not fairness.

Immediately retrain the model on recent data to adapt to the drift.

Why wrong: Retraining without analyzing the nature of drift may lead to suboptimal model.

Reduce the classification threshold to 0.5 to increase sensitivity.

Why wrong: Threshold change does not address feature drift.

Underfitting

Why wrong: Incorrect; underfitting would have low accuracy on both training and validation.

Data augmentation failure

Why wrong: Incorrect; data augmentation would affect training data diversity, not directly indicated.

Model compression

Why wrong: Incorrect; model compression reduces model size, not shown here.

Vanishing gradient

Why wrong: Vanishing gradient would cause training loss to plateau early, not the observed pattern.

Learning rate too high

Why wrong: A high learning rate would cause loss to oscillate or diverge rapidly, not a gradual decrease then increase in validation loss.

Underfitting

Why wrong: Underfitting would show high training and validation loss throughout, not the diverging pattern seen here.