medium · FRM Part 2 Current Issues
A bank uses a black-box neural network for credit limit decisions. To provide 'adverse action notices,' it employs a local surrogate model technique. A validator finds that for a single applicant, two different 'post-hoc' explainers provide conflicting reasons for rejection.
This phenomenon highlights which specific risk in XAI validation?
- Demographic parity concerns group-level fairness in outcomes across protected classes, not whether an explainer reflects the model's local logic.
- The lack of 'local fidelity,' where the surrogate model fails to accurately map the complex model's behavior in the specific region of that applicant's data.
- The exclusion fallacy describes a model reconstructing a banned attribute via proxy variables, unrelated to a validation tool's instability across post-hoc explainers.
- Overfitting concerns how well the model generalizes to unseen data, whereas conflicting explainers reflect instability in the post-hoc tools, not the model itself.
Sign up free to see the explanation and track your rank →
More FRM Part 2 Current Issues practice
- According to the BCBS standard for cryptoassets, a bank's to… — What is this limit?
- What primary balance (as a % of GDP) is required to stabilize the debt-to-GDP ratio?
- SVB's management removed interest rate hedges in 2022. According to the Economic Value of
- In monitoring model stability, which metric is commonly used to quantify the shift in dist
- What does a SHAP value of zero for a specific feature in a credit scoring model imply?
- A bank provides a 'NAV facility' to a private equity fund. If the fund's underlying assets
- What is the bank's primary ongoing obligation to regulators regarding this new model?
- If the firm's EBITDA declines, when can the lender intervene?