KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
CI/CD for ML
Continuous integration and continuous deployment pipeline specifically adapted to machine learning model lifecycles, integrating data validation, automated training, and controlled model deployment in production.
Automated Retraining Pipeline
Orchestrated workflow that automatically triggers model retraining based on predefined criteria (time-based, performance-based, or data drift), including validation and conditional deployment.
Shadow Deployment
Deployment strategy where the new model runs in parallel with the production model without affecting users, allowing silent performance validation before complete switchover.
Canary Deployment for ML
Gradual deployment approach where the new model is first exposed to a small percentage of traffic, with intensive monitoring before gradual extension to all requests.
ML Experiment Tracking
Structured logging system for hyperparameters, metrics, artifacts, and results of ML experiments, enabling systematic comparison and reproduction of training runs.
Continuous Model Evaluation
Automated process for continuous evaluation of model performance in production against reference test sets, including regression detection and bias metrics.
Model Governance Pipeline
Set of automated controls validating regulatory compliance, algorithmic fairness, and model documentation before their promotion to production.
Feature Engineering Automation
Automated pipeline for feature creation, transformation, and selection, with temporal stability validation and distribution drift tracking to maintain predictive quality.
ML Model Validation
Systematic step in the CI/CD pipeline that verifies model robustness, generalization, and compliance before deployment, including unit tests, integration tests, and business validation.
Hyperparameter Optimization CI
Continuous integration of hyperparameter optimization in the build pipeline, automating the search for optimal configurations with cross-validation and result tracking.
Model Explainability Pipeline
Automated workflow that generates and validates model explanations (SHAP, LIME) during CI, ensuring transparency and interpretability before production deployment.