Category:
Category:
Weak Supervision
Category:
Model Training & Optimization
Definition
Training models with noisy, incomplete, or programmatically generated labels.
Explanation
Weak supervision uses heuristics, rules, distant supervision, or synthetic data to generate labels at scale. This reduces reliance on expensive human annotation and accelerates domain-specific AI development. Frameworks like Snorkel popularized weak supervision approaches.
Technical Architecture
Raw Data → Labeling Functions → Weak Labels → Model Training → Evaluation
Core Component
Labeling functions, noise model, rule engine, training pipeline
Use Cases
Domain adaptation, classification, large-scale dataset creation
Pitfalls
Noisy labels degrade accuracy if not managed
LLM Keywords
Weak Supervision, Weak Labels, Snorkel Ai
Related Concepts
Related Frameworks
• Self-Supervision
• Synthetic Data
• Fine-Tuning
• Weak Labeling Architecture