Models are not complete products
Terms such as transformer, embedding and fine-tuning describe model capabilities. Grounding, guardrails, monitoring and human review describe the system around them.
Artificial Intelligence Glossary: 100+ AI and Machine Learning Terms
Understand 100+ artificial intelligence terms covering machine learning, deep learning, generative AI, NLP, computer vision, prompt engineering and AI agents.
A USEFUL MENTAL MODELLearn the language.
Learn the language.
Understand the system.
AI vocabulary stretches from mathematical model training to the design of complete systems that retrieve evidence, call tools and operate under human oversight. This glossary separates those layers so you can recognise whether a term describes data, a model, an evaluation method, an interface or an operational control.
Evaluation needs context
Accuracy, precision and recall answer different questions. A useful metric must reflect the consequence of each kind of error.
Responsible use is technical work
Safety, governance and privacy belong in architecture and evaluation—not only in a policy written after deployment.
FROM DEFINITION TO DIRECTIONEvery term now connects
Every term now connects
to what to learn next.
See exactly where the concept appears in a structured specialisation.
Read the roadmap, comparison or project guide that gives it context.
Connect relevant terminology with the certification domain that assesses it.
108 termsClear definitions · Relevant learning paths
Accuracy
The proportion of evaluated predictions that a model classifies correctly across all classes.
Activation function
A mathematical function that introduces non-linearity into a neural network layer.
Agent orchestration
The coordination of model calls, tools, state, checks and handoffs inside an agent system.
Agentic workflow
A controlled sequence in which an AI system selects tools and actions to pursue a goal.
AI agent
A software system that uses a model, tools and state to pursue a goal through one or more controlled actions.
AI alignment
The practice of shaping an AI system so its behaviour remains consistent with intended human goals and constraints.
AI assistant
An AI application designed to help a person complete defined information, creation or decision-support tasks.
AI ethics
The study and practice of developing AI with attention to fairness, accountability, rights and social impact.
AI governance
Policies, responsibilities and controls used to oversee AI systems throughout their lifecycle.
AI safety
Methods for preventing, detecting and reducing harmful or unintended behaviour in AI systems.
Anomaly detection
The task of identifying observations that differ meaningfully from expected patterns.
Artificial intelligence
The field of designing systems that perform tasks involving learning, reasoning, perception, language, planning or decision-making.
Attention mechanism
A neural-network mechanism that weighs which parts of an input are most relevant while producing a representation or output.
Autonomous system
A system that can select and execute actions with limited immediate human direction.
Autoregressive model
A model that predicts the next element using previously observed or generated elements.
Backpropagation
The algorithm used to calculate how neural-network parameters contributed to an error during training.
Batch
A group of training examples processed together before a model update.
Beam search
A decoding method that keeps several promising output sequences while generating text or other sequences.
Bias
A systematic tendency in data, modelling or evaluation that can distort outcomes for particular situations or groups.
Chain-of-thought prompting
A prompting approach that encourages intermediate reasoning before a final answer, where appropriate and safely exposed.
Chatbot
A conversational interface that exchanges messages with users through rules, models or both.
Chunking
The division of documents into retrievable segments that fit a model's context and preserve useful meaning.
Classification
A modelling task that assigns an input to one or more predefined categories.
Clustering
An unsupervised learning task that groups observations according to measured similarity.
Computer vision
The field of building systems that extract useful information from images or video.
Confusion matrix
A table comparing predicted and actual classes to reveal different kinds of classification error.
Context engineering
The design of instructions, retrieved evidence, state and tools supplied to a model at runtime.
Context window
The maximum amount of input and generated information a model can consider in one interaction.
Convolutional neural network (CNN)
A neural-network architecture that learns spatial patterns and is widely used for image tasks.
Cosine similarity
A measure of the angle between vectors, often used to compare embeddings.
Cross-validation
A resampling method that evaluates a model across multiple train-and-validation splits.
Data drift
A change in real-world input patterns that can reduce a deployed model's reliability.
Deep learning
A branch of machine learning that uses multilayer neural networks to learn representations from data.
Diffusion model
A generative model that learns to create data by reversing a gradual noising process.
Dimensionality reduction
The transformation of many features into fewer informative dimensions while retaining useful structure.
Discriminative model
A model that learns boundaries or conditional relationships for predicting an output from an input.
Embedding
A numerical representation that places related items near one another in a learned vector space.
Epoch
One complete pass through the available training dataset during model training.
Evaluation dataset
A held-out collection used to measure model behaviour on representative tasks and edge cases.
Explainable AI (XAI)
Methods that help people understand the factors, evidence or reasoning behind an AI output.
False negative
A case where a model incorrectly predicts that a condition or class is absent.
False positive
A case where a model incorrectly predicts that a condition or class is present.
Feature
A measurable input or representation used by a machine-learning model to make a prediction.
Few-shot learning
Performing or adapting to a task using only a small number of examples.
Fine-tuning
Additional model training on selected data to adapt behaviour to a narrower task or domain.
Foundation model
A broadly trained model that can be adapted or prompted for many downstream tasks.
Generative AI
AI systems that create or transform content such as text, code, images, audio or structured data.
Gradient descent
An optimisation method that repeatedly adjusts parameters to reduce a model's loss.
Grounding
Connecting a model response to supplied evidence, data or tool results rather than unsupported recall.
Guardrail
A technical or procedural control that constrains, checks or redirects AI behaviour.
Hallucination
A fluent model output that is unsupported, invented or inconsistent with the available evidence.
Hallucination rate
The measured frequency of unsupported or invented claims across a defined evaluation set.
Human in the loop
A workflow that requires human review, feedback or approval at selected decision points.
Hyperparameter
A training or architecture setting chosen outside the model's learned parameters.
Image segmentation
A computer-vision task that assigns a category or identity to pixels or image regions.
Inference
The process of using a trained model to produce a prediction, representation or generated output.
Instruction tuning
Additional training that improves a model's ability to follow natural-language requests and examples.
Knowledge graph
A structured network of entities and relationships used to represent connected knowledge.
Label
The known target category or value attached to a supervised-learning example.
Large language model (LLM)
A model trained on large text collections to predict and generate language across many tasks.
Latent space
A learned numerical space where a model represents underlying features or concepts.
Learning rate
The step size used when updating model parameters during optimisation.
Linear regression
A statistical learning method that models a numerical outcome as a weighted combination of inputs.
Loss function
A mathematical measure of the difference between a model's output and the desired target.
Machine learning
Methods that learn patterns from data to make predictions, representations or decisions.
Model card
Documentation describing a model's purpose, evaluation, limitations, data context and responsible-use considerations.
Model evaluation
The systematic measurement of model behaviour on representative data using task-relevant criteria.
Model monitoring
The ongoing measurement of deployed model quality, drift, latency, safety and operational behaviour.
Model quantisation
Reducing numerical precision to decrease model memory or compute requirements while managing quality loss.
Multimodal model
A model that can process or generate more than one modality, such as text and images.
Named entity recognition
An NLP task that identifies and classifies names, organisations, places and other entities in text.
Natural language processing (NLP)
The field of building systems that analyse, understand or generate human language.
Neural network
A layered computational model that learns parameters connecting inputs to representations or outputs.
Object detection
A computer-vision task that locates and classifies objects within an image or video frame.
One-shot learning
Learning or performing a task from a single labelled example or demonstration.
Overfitting
When a model learns training examples too specifically and performs poorly on new representative data.
Parameter
A numerical value learned during training that influences how a model transforms inputs into outputs.
Precision
Among predicted positive cases, the proportion that are actually positive.
Pretraining
Initial model training on broad data before adaptation to narrower tasks.
Prompt engineering
The disciplined design, testing and maintenance of instructions and context for generative-model tasks.
Prompt injection
An attempt to manipulate a model through instructions embedded in user input or retrieved content.
Random forest
An ensemble method that combines many decision trees to improve predictive stability.
Recall
Among actual positive cases, the proportion that a model correctly identifies.
Red teaming
Structured adversarial testing intended to expose unsafe, unreliable or exploitable AI behaviour.
Reinforcement learning
Learning behaviour through actions, feedback and rewards within an environment.
Responsible AI
The practice of designing and operating AI with safety, fairness, privacy and accountability.
Retrieval-augmented generation (RAG)
A pattern that retrieves relevant source material and supplies it to a generative model as grounding context.
Reward model
A model trained to estimate how well another model's output matches desired preferences or goals.
Safety filter
A control that detects or blocks content and actions outside defined safety policies.
Semantic search
Retrieval based on the meaning of a query and content rather than exact keyword matching alone.
Sentiment analysis
The classification of attitudes or emotional polarity expressed in text.
Speech recognition
The conversion of spoken audio into text or structured language representations.
Supervised learning
Machine learning from examples that pair inputs with known target labels or values.
Synthetic data
Artificially generated examples designed to supplement, simulate or protect real data.
Temperature
A generation setting that controls how concentrated or varied a model's token choices are.
Text classification
The assignment of documents or messages to predefined language categories.
Text-to-image model
A generative model that creates images from natural-language descriptions.
Token
A unit of text processed by a language model, which may be a word, part of a word or punctuation.
Tool calling
A model's structured request for an external function, API or service to perform an action.
Top-p sampling
A decoding method that samples from the smallest token set containing a chosen probability mass.
Training data
The examples used by a learning algorithm to estimate model parameters.
Transfer learning
Reusing representations learned for one task or dataset to improve another.
Transformer
A neural-network architecture built around attention, widely used for language and other sequence or representation tasks.
Unsupervised learning
Learning patterns or structure from data without provided target labels.
Validation data
Held-out examples used to choose model settings without using the final test set.
Vector database
A database designed to store and search numerical embeddings efficiently.
Vision-language model
A multimodal model that connects visual inputs with natural-language understanding or generation.
Zero-shot learning
Performing a task without task-specific examples by using learned representations or instructions.
Questions about artificial intelligence vocabulary.
What AI terms should a beginner learn first?+
Begin with artificial intelligence, machine learning, training data, feature, model, inference, evaluation and overfitting. Then add terms from the application area you want to build.
Are generative AI and machine learning the same?+
Generative AI is an application family built using machine-learning models. Machine learning also covers prediction, classification, clustering and many non-generative tasks.
How should I use this AI glossary?+
Read the definition, follow its related Certificate of Specialisation, and connect the term to a project decision or evaluation example.
FROM DEFINITION TO PRACTICE
Related Certificate of Specialisation programmes.
Use the glossary to clarify the language, then inspect the complete 30-hour syllabus and projects for your chosen technology.
AI01
Artificial Intelligence
Design, evaluate and responsibly apply modern intelligent systems.
🎁 Career Edge Included FREE
₹3,150Complete programme
↗
ML02
Machine Learning
Build, evaluate and improve end-to-end predictive models.
🎁 Career Edge Included FREE
₹3,150Complete programme
↗
GA03
Generative AI
Build grounded, evaluated applications with modern generative AI.
🎁 Career Edge Included FREE
₹3,150Complete programme
↗
AG04
AI Agents
Create tool-using agents with memory, control and measurable reliability.
🎁 Career Edge Included FREE
₹3,150Complete programme
↗