AI & machine
learning glossary.
30 terms explained the way a working engineer would explain them — what it is, how it actually works, and when it is the wrong tool. No marketing definitions.
Foundations terms.
Confusion Matrix
A confusion matrix is a table of a classifier's predictions against actual outcomes, splitting results into true positives, false positives, true negatives, and false negatives so that different kinds of error can be measured separately.
Hyperparameter Tuning
Hyperparameter tuning is the search for configuration values that are set before training rather than learned during it, such as learning rate, model depth, and regularisation strength.
Inference
Inference is the act of running a trained model on new input to produce a prediction, and because it happens on every request it usually dominates the total lifetime cost of a machine learning system.
Neural Network
A neural network is a machine learning model built from layers of simple weighted units that learn to map inputs to outputs by repeatedly adjusting those weights to reduce prediction error.
Overfitting
Overfitting is when a model fits the particular noise and quirks of its training data so closely that its performance on new, unseen data gets worse rather than better.
Reinforcement Learning
Reinforcement learning trains an agent to choose actions in an environment by rewarding desirable outcomes, so the agent learns a policy that maximises cumulative reward over time.
Supervised Learning
Supervised learning is training a model on examples that already carry the correct answer, so the model learns a mapping it can apply to new, unlabelled inputs.
Training Data
Training data is the set of examples a model learns from, and its coverage, label quality, and representativeness set a hard ceiling on how well any model trained on it can perform.
Transfer Learning
Transfer learning takes a model already trained on a large general dataset and adapts it to a narrower task, so the new task needs far less labelled data and compute than training from scratch.
Unsupervised Learning
Unsupervised learning finds structure in data that has no labels, producing groupings, compressed representations, or density estimates rather than predictions of a known target.
Language Models terms.
AI Agent
An AI agent is a language model placed in a loop with access to external tools, so that it can decide on an action, observe the result, and iterate toward a goal rather than producing a single response.
Embedding
An embedding is a list of numbers representing a piece of content in a space where geometric closeness corresponds to semantic similarity, so that meaning becomes something you can measure arithmetically.
Fine-Tuning
Fine-tuning continues training a pretrained model on a smaller task-specific dataset so that its weights adapt to your format, tone, and domain conventions.
Knowledge Graph
A knowledge graph represents information as entities connected by typed, directional relationships, so that connections between things are first-class data that can be queried and traversed.
Large Language Model
A large language model is a neural network trained on enormous volumes of text to predict the next token, which turns out to produce fluent language, broad factual recall, and useful reasoning as side effects.
Prompt Engineering
Prompt engineering is the practice of designing the text supplied to a language model — instructions, examples, and context — so that its outputs are accurate, correctly formatted, and consistent enough to build on.
Quantisation
Quantisation reduces the numerical precision used to store a model's weights and activations, shrinking memory footprint and speeding up inference at a usually small cost in accuracy.
Retrieval-Augmented Generation
Retrieval-augmented generation fetches relevant passages from your own corpus at query time and supplies them to a language model as context, so answers are grounded in specific sources rather than in the model's frozen training data.
Transformer
A transformer is a neural network architecture that processes a whole sequence at once using an attention mechanism, letting every position directly consult every other position instead of passing information along step by step.
Vector Database
A vector database stores high-dimensional embeddings and retrieves the nearest ones to a query vector in sub-linear time, using approximate nearest-neighbour indexes rather than comparing against every record.
Production & Applications terms.
Anomaly Detection
Anomaly detection identifies observations that deviate meaningfully from an established pattern of normal behaviour, which makes it suitable for problems where the failures you care about are rare and not known in advance.
Computer Vision
Computer vision is the field concerned with extracting structured information from images and video, covering tasks from whole-image classification to pixel-level segmentation and object tracking.
Edge AI
Edge AI runs model inference on local hardware — a phone, camera, vehicle, or industrial controller — rather than sending data to a remote server for processing.
Explainable AI
Explainable AI covers the methods used to make a model's decisions inspectable by humans, either by choosing an inherently interpretable model or by generating post-hoc explanations of a complex one.
Feature Store
A feature store is infrastructure that computes, stores, and serves model input features from a single definition, so that training and live serving use identical logic and history is reconstructed correctly.
MLOps
MLOps is the set of practices for deploying, monitoring, and maintaining machine learning models in production, extending DevOps to cover the parts of a system that change because data changes rather than because code changes.
Model Drift
Model drift is the gradual loss of a deployed model's accuracy because the statistical relationship it learned no longer matches the data or the world it now operates in.
Natural Language Processing
Natural language processing is the field concerned with getting computers to work with human language, spanning classification, information extraction, search, translation, and generation.
Recommendation Engine
A recommendation engine predicts which items a particular user is most likely to want, ranking a catalogue against that user's history and context rather than showing everyone the same thing.
Synthetic Data
Synthetic data is artificially generated data that mimics the statistical properties of real data, used when real data is scarce, restricted, dangerously imbalanced, or legally difficult to share.
Turning theory into
a shipped system.
We design, train, and deploy production machine learning. Tell us what you are trying to build.
Start the conversationor write to us at [email protected]