- Learn
- Artificial Intelligence
- Representations: Embeddings and Vector Spaces
AI-M04
Representations: Embeddings and Vector Spaces
How modern AI turns words, sentences, images and users into vectors, so that similarity of meaning becomes distance you can compute and search.
Search, logic and probability all worked with symbols: a word was a name, and two different names had nothing in common. Modern AI works with embeddings instead, lists of numbers placed so that things with similar meaning sit close together. Once meaning is geometry, "find something like this" becomes "find the nearest point", and that single idea powers semantic search, recommendations, retrieval for language models and much more.
This short module builds the idea from the ground up. You will see why one-hot symbols cannot express similarity and how dense vectors can, then meet embedding models that turn whole sentences and documents into a single vector, compare them with cosine similarity, and weigh dimensions, cost and language support when choosing one. Finally you will build semantic search in its simplest form and learn when plain keyword search still wins, the foundation for retrieval augmented generation in AI-M08.
Lessons
- AI-04.01 From symbols to vectors Symbols such as one-hot vectors treat every word as unrelated to every other. Beginner 10 min
- AI-04.02 Sentence and document embeddings Embedding models turn whole sentences or documents into a single vector that captures their meaning. Intermediate 12 min
- AI-04.03 Semantic search basics Keyword search matches exact words; semantic search matches meaning by embedding the query and finding the nearest document embeddings. Beginner 10 min