Production Machine Learning Monitoring: Outliers, Drift, Explainers and Statistical Performance

Dive into advanced techniques for monitoring machine learning models in production environments. Explore best practices, principles, patterns, and techniques for effective production monitoring of ML models. Cover standard microservice monitoring methods applied to deployed models, as well as advanced paradigms like concept drift, outlier detection, and explainability. Follow a hands-on example of training an image classification model from scratch, deploying it as a microservice in Kubernetes, and implementing advanced monitoring components. Learn about AI explainers, outlier detectors, concept drift detectors, and adversarial detectors. Understand high-level architectural patterns that abstract complex monitoring techniques into scalable infrastructural components, enabling monitoring across numerous heterogeneous ML models. Gain insights from Alejandro Saucedo, Engineering Director at Seldon, on standardized interfaces and best practices for managing the lifecycle of machine learning models in production.