Loss Function
A mathematical function that quantifies the difference between predicted outputs and true values, guiding model training and optimization.
Definition
The objective metric (e.g., cross-entropy, mean squared error, custom fairness-penalty losses) that the optimizer minimizes. Choice of loss function directly influences model behavior—governance must review loss definitions to ensure alignment with business goals, ethical constraints (e.g., adding fairness regularizers), and risk tolerance before training, and must document configurations for auditability and reproducibility.
Real-World Example
A fraud-detection model’s team chooses a weighted cross-entropy loss that penalizes false negatives four times more than false positives—reflecting the business cost of missed fraud. They document the weighting rationale and track downstream impact on both precision and recall to ensure balanced outcomes.
.svg)
