What it is: A method of teaching computers to recognize patterns and make predictions from data, without explicitly programming every rule.
Key types for vectors:
- Supervised Learning: Learn from labeled examples (e.g., “this is a cat” → cat embedding)
- Unsupervised Learning: Find patterns in unlabeled data
- Neural Networks: Layered systems that learn complex patterns (how most embeddings are created)
Why it matters for vectors: ML models create the embeddings that vector databases store and search. Better ML models → better embeddings → more accurate search results.
Simple analogy: Like teaching a child to recognize animals by showing them thousands of pictures with labels, until they can identify new animals they’ve never seen before.