Concept Drift

TECHNICAL DEFINITION

Concept drift is a specific type of model drift where the underlying relationship between the input features and the target variable changes over time, meaning the

BACKGROUND

Adversarial machine learning is the study of the attacks on machine learning algorithms, and of the defenses against such attacks.

SYNONYMS & ALIASES

• Target drift
• label drift
• semantic drift

USAGE NOTE

Concept drift is particularly challenging in dynamic environments like fraud detection or recommendation systems.

DEVELOPERS

Organizations developing technology related to Concept Drift.

• Arize AI

  Arize AI offers an ML observability platform that helps data scientists and ML engineers monitor, troubleshoot, and explain models in production. A core capability is detecting and diagnosing concept drift and data drift, ensuring models maintain performance over time.

• WhyLabs AI

  WhyLabs AI provides an AI observability platform called 'whylogs' that helps track, monitor, and debug data pipelines and ML models. It specializes in detecting data drift, concept drift, and data quality issues in production, crucial for maintaining model integrity.

• Fiddler AI

  Fiddler AI delivers an Explainable AI (XAI) and ML Monitoring platform that enables enterprises to understand, validate, and monitor their AI models. It actively helps detect performance degradation caused by concept drift and provides tools for diagnosis.

• Amazon Web Services (AWS) Sagemaker

  AWS SageMaker offers a comprehensive suite of machine learning services, including robust model monitoring capabilities. Its SageMaker Model Monitor can detect data drift and concept drift in production models, alerting users to performance degradation and enabling timely retraining.

• Google Cloud AI (Vertex AI)

  Google Cloud's Vertex AI provides a unified platform for building, deploying, and scaling ML models. Its Model Monitoring features are designed to detect data drift, concept drift, and anomaly detection, helping maintain model accuracy and reliability in production.

• Microsoft Azure Machine Learning

  Azure Machine Learning provides an end-to-end platform for the ML lifecycle, including capabilities for model monitoring. It offers tools for data drift detection and concept drift analysis, helping users understand when model retraining is necessary due to changing data patterns.

• Databricks

  Databricks offers a unified data and AI platform that includes MLflow for MLOps. Its capabilities support monitoring deployed models, detecting performance issues, and managing model retraining workflows that are essential for addressing concept drift in production environments.

• Seldon

  Seldon provides open-source and enterprise MLOps platforms (Seldon Core, Seldon Deploy) for deploying, managing, and monitoring machine learning models at scale. Their monitoring capabilities focus on detecting model performance degradation and data drift, which are key indicators of concept drift.