DataTerminology.comNEW
Also known as: CT
Regularly retraining models with new data to maintain performance.
DESCRIPTION
Continuous training automates model retraining triggered by new data, drift detection, or schedule. It integrates data validation, feature engineering, training, evaluation, and promotion. CI/CD principles apply with pipelines, versioning, and reproducibility. CT reduces staleness and adapts models to changing environments. It requires strong monitoring, governance, and rollback protections for safe production updates.
ALTERNATIVE DEFINITIONS
Google Cloud
Continuous training (CT) is the process of automatically retraining models on new data and deploying updated models in production.
Stresses automation and production deployment of updates.
TensorFlow Extended (TFX)
TFX pipelines enable continuous training by orchestrating data validation, transformation, training, and evaluation as new data arrives.
Highlights pipeline orchestration enabling CT with data-aware components.
RELATED TERMS
MLOps Continuous Integration Continuous Deployment Model Monitoring Data Drift Pipelines Feature Store Retraining Model Registry Model Serving
COMPETING TERMS
Batch Training
RELATIONS
- Broader: Training
KNOWLEDGE AREAS
MLOps AI Data Science Observability & Telemetry Testing & Validation
TARGET AUDIENCE
MLOps Engineer Data Scientist AI Engineer Data Engineer
EXTERNAL SOURCES
- Wikipedia - MLOps
- Google Cloud - CT patterns and architecture
- Kubeflow - Pipelines for ML automation
- AWS - SageMaker pipelines for CT
TOP QUESTIONS
Why is Continuous Training important?
Data distributions change; CT keeps models current, reducing performance degradation and bias accumulation over time.
Category: Relevancy
What triggers Continuous Training?
Schedule, drift thresholds, data volume, quality signals, or business events can trigger retraining.
Category: Requirements
What are common mistakes with Continuous Training?
Skipping data validation, inadequate evaluation baselines, and promoting models without governance increase risk.
Category: Mistakes
What risks exist in Continuous Training?
Unintended regressions, feedback loops, and cost overruns; strong monitoring and rollback are essential.
Category: Risk
How is Continuous Training managed operationally?
Automate pipelines, use model registries, enforce approvals, and monitor metrics for drift and performance.
Category: Management
What is a hot topic around Continuous Training?
Adaptive retraining with active learning and policy-based promotion using SLOs and guardrails.
Category: Topic
How does Continuous Training relate to CI/CD?
CT extends CI/CD to data and models, ensuring repeatable, tested updates with versioned artifacts.
Category: Example
Where should teams start with Continuous Training?
Define drift metrics, build a simple pipeline, integrate validation, and add promotion gates.
Category: Requirements
What is the alternative to Continuous Training?
Batch Training retrains periodically regardless of drift, which may be simpler but less responsive.
Category: Alternatives