Embedding
A numerical representation of data that captures semantic meaning in a high-dimensional vector space.
Detailed Definition
An embedding is a dense, numerical representation of data (such as words, sentences, images, or other objects) in a high-dimensional vector space, where similar items are positioned close to each other. Embeddings are fundamental to modern AI systems because they allow machines to understand and work with semantic relationships. For example, word embeddings represent words as vectors where synonyms and related words cluster together in the vector space. This enables AI systems to understand that 'king' and 'monarch' are related, or that 'dog' and 'puppy' share semantic similarity. Embeddings are created through training neural networks on large datasets, learning to encode meaningful relationships and patterns. They're crucial for applications like search engines (semantic search), recommendation systems, and retrieval-augmented generation (RAG) systems. Modern embedding models can capture complex relationships and enable AI systems to perform sophisticated reasoning about similarity and relevance.
Core TechnologiesMore in this Category
Autoregressive Model
A type of model that predicts the next element in a sequence based on previous elements.
BERT
Bidirectional Encoder Representations from Transformers - a pre-trained language model.
Deep Learning
A subset of machine learning using neural networks with multiple layers to learn complex patterns.
GPT (Generative Pre-trained Transformer)
A family of language models that generate human-like text using transformer architecture.
Transformer
The neural-network architecture that introduced ‘attention’ and enabled today’s LLM revolution.