Embeddings in Machine Learning

What are Embeddings in Machine Learning?

Embeddings are dense vector representations of discrete objects—words, documents, users, images—learned so semantic or behavioral similarity corresponds to geometric closeness.

They power search, recommendations, and RAG. Weak or poisoned embeddings undermine retrieval quality and can amplify security issues in LLM pipelines.

Related reading on Giskard

• LLM hallucinations and misinformation detection
• Practical guide on AI security and LLM vulnerabilities

Stress-test embedding-backed retrieval — Weak or poisoned embeddings undermine RAG—pair similarity metrics with groundedness and injection probes on the full pipeline. Learn more: LLM hallucinations and misinformation detection.

Authoritative reference: Sentence-Transformers