L1 L2 Regularization
Domain: AI Fundamentals
🎯 Motivation
L1 adds a penalty based on absolute weight values (can zero out weights). L2 adds a penalty based on squared weights.
📋 Concrete Example
L1 can shrink unimportant weights to exactly zero, doing feature selection. L2 keeps all weights small but nonzero.
🔗 Analogy
L1 is like a strict editor who deletes whole paragraphs. L2 is like one who trims every word a little.