AI & ML Tech Glossary

Model-as-a-Judge is an evaluation approach where a language model grades another model’s outputs using a rubric, returning scores, rankings, or critiques. It enables scalable automated...

Multimodal RAG extends retrieval-augmented generation to text plus assets like images, tables, and charts, retrieving relevant multimodal evidence and grounding a multimodal model’s answers to...

Model distillation trains a smaller student model to mimic a larger teacher model by matching its outputs or representations, often reducing latency and cost while...

Model parallelism splits a single neural network across multiple GPUs or accelerators so each device computes part of the model. It enables training and serving...

A model stealing attack uses repeated queries to a deployed ML/LLM (often via API) to train a surrogate that mimics the victim model, potentially exposing...

Model Context Protocol (MCP) standardizes how AI apps connect to tools and data sources, letting agents discover and invoke capabilities through a consistent, governed interface...

Multimodal retrieval searches across text, images, audio, or video using embeddings that support cross-modal matching, such as finding images from a text query or documents...

Mixture of Experts (MoE) is a sparse neural architecture where a router selects a small number of expert sub-networks per token. This boosts total model...