AI 용어집
인공지능 완전 사전
Automated Model Retraining
Systematic process of updating machine learning models in production, triggered by performance metrics or data changes, without manual intervention.
Continuous Model Monitoring
Real-time monitoring of performance metrics, data drift, and prediction behavior to ensure the reliability and relevance of deployed models.
Automated Feature Engineering
Automatic generation and selection of optimal predictive variables from raw data, using algorithms to create relevant transformations.
Hyperparameter Optimization (HPO)
Automated search for the best hyperparameters for a given model, using techniques like grid search, random search, or Bayesian optimization.
Model Explainability Automation
Automatic generation of model prediction interpretations using techniques like SHAP or LIME to ensure transparency and trust.
Automated Data Validation
Systematic verification of input data quality and compliance against a schema or reference statistics before model usage.
Canary Deployment for ML Models
Progressive deployment strategy where a new model version is tested on a small subset of traffic before full deployment to minimize risks.
Automated A/B Testing for Models
Automatic setup of comparative experiments between different model versions to statistically evaluate their performance in real conditions.
Model Versioning Automation
Systematic management of different model versions, their metadata, and associated artifacts to ensure traceability and reproducibility.
Automated Model Packaging
Automatic process of preparing models for deployment, including serialization, API creation, and dependency configuration.
Resource Auto-scaling for ML Inference
Dynamic and automatic adjustment of computing resources (CPU, GPU, memory) based on prediction load to optimize costs and performance.
Automated Model Governance
Systematic implementation of policies, audits, and documentation to ensure regulatory compliance and ethics of automated models.
Automated Pipeline Orchestration
Automatic coordination of all ML lifecycle steps, from data ingestion to production monitoring, through defined workflows.
Model Performance Degradation Alerting
Automatic notification system triggered when key model metrics fall below predefined thresholds, indicating a need for intervention.