Metrics: Membership inference attack
Overview
H2O Model Security offers an array of metrics in the form of a table and a statistics card to understand a membership inference attack.
Membership inference attack table
| Column name | Column description |
|---|---|
| Estimated training | The Estimated training refers to the prediction of the second-level surrogate model attributed to the row. In other words, it indicates whether the second-level surrogate model believes H2O Model Security utilized the row as training data for the first-level surrogate model. A 0 indicates the row was not in the first-level surrogate model, and 1 indicates otherwise. |
| Probability in training | The Probability in training refers to the second-level surrogate model's probability for the prediction being 0 or 1, indicated in the Estimated training column. |
| Actual training value | The Actual training value indicates whether the row was in the first-level surrogate model as training data. A 0 indicates the row was not in the first-level surrogate model, and 1 indicates otherwise. |
| ... (rest of columns) | The other columns in the table represent the features required to score new data. In other words, represent the data (columns) needed to obtain a prediction. |
Accuracy of in_train prediction
The Accuracy of in_train prediction stats card displays the number of rows correctly labeled as part of the first-level surrogate model's training dataset. As well, H2O Model Security displays the number of correct predictions as a percentage value out of 100%.
Feedback
- Submit and view feedback for this page
- Send feedback about H2O Model Security to cloud-feedback@h2o.ai