Machine Learning Architecture

Kind: model

Layer: Correctness Core

Record: architecture:machine-learning-architecture

Severity: contextual

Scope: ML pipeline, model serving, data

Canonical: Ontology

A conceptual representation of a model's lifecycle as versioned data, feature, training, evaluation and serving stages.

Listed in Architecture principles, after Artificial Intelligence Architecture and before Model Governance.

Requires

• Data Pipeline
• Training/Inference Separation

Reinforces

• Reproducibility
• Model Governance

Enables

• Reliable Model Lifecycle

Conflicts with

• Ad-Hoc Notebook-to-Production

In tension with

• Experimentation Speed

Tensions

• Machine Learning Architecture / Experimentation Speed

Violated by

• Model Version Ambiguity

Refactored by

• Data Pipeline
• Prompt Versioning

Severity

• contextual

Category

• Model Architecture

Linked from

• Correctness / Determinism / Verification
• Model Architecture
• Model Architecture
• Severity levels
• The resolutions