Organized by AI Model Lifecycle Phase

Compiled by Lubos, Edge AI Auditor & Open Compliance Advisor

Version 1.1 – June 2025

🧩 1. Data Collection

These terms relate to how and what data you collect to train or operate AI models — especially important for GDPR and AI Act compliance.

• Data Minimization – Collect only what’s necessary
• Provenance – Know the origin of your training data
• Synthetic Data – Fake but useful data for testing AI
• Differential Privacy – Mask real user data mathematically

🧠 2. Model Development

These cover how the AI is built — including performance, fairness, and readiness for edge devices.

• Model Compression – Shrinking models for efficiency
• Lightweight Model – Designed for small devices
• Bias Detection – Test for fairness across user groups
• Model Card – Document model behavior and usage limits

🧪 3. Evaluation & Testing

What happens before launch — proving your model is safe and responsible.

• Explainability – Can a human understand the model’s output?
• Risk Scoring – Estimate the harm a model might cause
• Sandbox Mode – Test AI in a safe, isolated environment