EqualizedOddsChecker<T>
Models & Types · AiDotNet.Safety.Fairness
Checks for equalized odds violations by analyzing whether model quality or effort varies across demographic groups mentioned in the text.
Implements: ITextSafetyModule<T>
For Beginners
Imagine asking an AI to write a recommendation letter for two equally qualified candidates. If the letter for a man is detailed and enthusiastic but the letter for a woman is shorter and uses more hedging words like “might” or “could”, that’s an equalized odds violation — the model is giving different quality output for different groups.
How It Works
Equalized odds requires that the true positive rate and false positive rate are equal across all demographic groups. Since we’re working with text (not tabular classification), this module detects differential quality indicators: response length disparity, detail level differences, hedging language, and conditional/qualifying statements that differ by demographic group.
References:
- Equalized Odds in Machine Learning (Hardt et al., NeurIPS 2016)
- Measuring algorithmic fairness in text generation (2024)
- BEATS: Comprehensive bias evaluation test suite for LLMs (2025)
Constructors
| Constructor | Summary |
|---|---|
EqualizedOddsChecker(double disparityThreshold) |
Initializes a new equalized odds checker. |
Properties
| Property | Summary |
|---|---|
bool IsReady { get; } |
|
string ModuleName { get; } |
Methods
| Method | Summary |
|---|---|
IReadOnlyList<SafetyFinding> Evaluate(Vector<T> content) |
|
IReadOnlyList<SafetyFinding> EvaluateText(string text) |