AI Glossary

Abductive reasoning involves taking observations or evidence and making the best possible hypothesis to explain them.

A type of AI focused on specific tasks like weather updates, data analytics, or gameplay, not general intelligence.

A specialized language used to define the actions an AI agent can perform and their consequences.

A formal language for representing and reasoning about actions in a changing environment.

AML is a method in AI for learning preconditions and effects of actions, often in planning systems.

Finding the best sequence of actions to achieve a specific goal efficiently.

The AI chooses the next action based on its goals and current context.

In neural networks, it determines whether a neuron fires based on input, shaping model output.

A learning approach where AI requests more data or examples to improve its performance.

A reinforcement learning technique where 'actor' selects actions and 'critic' evaluates them.

An algorithm that adjusts its behavior dynamically based on incoming data.

A search algorithm concept that ensures estimated cost is never overestimated for optimality.

The study of defending AI from attacks that attempt to mislead or trick learning models.

A technique in which AI anticipates opponent moves to find a winning strategy.

A software system that perceives its environment and takes actions to achieve goals.

The structural design of an AI agent, including sensors, decision units, and actuators.

A simulation approach that studies the behavior and interaction of autonomous agents.

Specialized hardware designed to speed up computations in AI applications.

Conversational AI tools powered by LLMs that assist users in completing organizational tasks.

Ensuring that AI systems behave in ways that are aligned with human values and ethics.

Creative artworks generated with the assistance of artificial intelligence.

A hypothetical containment setup to prevent powerful AI from affecting the real world.

A class of problems so complex that solving them would mean achieving true general AI.

The challenge of ensuring that superintelligent AI systems remain under human control.

The idea that once AI performs a task well, it’s no longer seen as intelligent.

A field of research ensuring AI technologies remain safe, ethical, and aligned with human benefit.

Software extensions that integrate AI functionality into existing applications.

A systematic error in an AI’s output leading to unfair or prejudiced outcomes.

A measure of how effectively an algorithm utilizes resources like time and memory.

A pioneering AI program by DeepMind that defeated human Go champions using deep learning.

A general-purpose AI by DeepMind that mastered multiple games from scratch without human data.

AI integrated into environments, creating seamless and intuitive user experiences.

The study of algorithm efficiency, including time complexity and resource consumption.

An algorithm that can return a valid solution even if interrupted before completion.

AI that can perform any intellectual task that a human can do.

AI modeled after the human immune system for detecting threats like cyberattacks.

The simulation or creation of life-like systems using computational models.

AI systems capable of making decisions based on ethical considerations.

A method where AI mimics brain-like processing to interpret and analyze data.

AI systems designed for autonomous decision-making and task execution with minimal human input.

Backpropagation is a training method where neural networks learn from errors by adjusting weights after each iteration.

An extension of backpropagation used to train recurrent neural networks that process sequential data.

A reasoning technique that starts from the goal and works backward to identify necessary conditions.

The simplest instance of a problem used to terminate recursive functions or algorithms.

A simple reference model used to compare the performance of more complex machine learning models.

A technique to improve neural network training by normalizing layer inputs, boosting stability and speed.

A classifier that uses Bayes' Theorem to predict the probability of classes given the input features.

The lowest possible classification error for a model given its features and the underlying data.

A standard dataset and metrics used to evaluate and compare the performance of machine learning models.

The tendency of an AI model to produce skewed results based on the data it was trained on.

A balance between a model’s ability to minimize error on training data and generalize to new data.

Bidirectional Encoder Representations from Transformers – a powerful NLP model pre-trained on vast text corpora.

A basic data type in programming that can hold only two values: true or false.

Extremely large and complex datasets used in AI for training and discovering patterns at scale.

A type of neural network that captures spatial relationships between parts of an object, enabling better recognition despite viewpoint changes.

A problem-solving method where new issues are solved by adapting solutions from similar past cases.

When a neural network forgets previous knowledge while learning new data, often in continual learning scenarios.

AI software that simulates human conversation, typically via text or voice, to complete tasks or provide information.

A generative AI chatbot using large language models to generate human-like responses across a wide range of prompts.

A supervised learning task where the model predicts category labels based on input features.

An unsupervised learning technique where similar data points are grouped together without labeled outputs.

Grouping similar data in datasets with many features, often using dimensionality reduction or advanced algorithms.

A blueprint for simulating human cognition, used in AI to build systems that mimic human thinking.

An internal representation used by AI agents to understand and navigate environments and make decisions.

A field focused on mimicking human thought processes using AI to learn and make decisions intelligently.

A linguistic term in NLP where a pronoun refers to something introduced later in the text.

Using AI to generate creative works such as art, music, literature, and design.

A theoretical study of machine learning algorithms and their performance, accuracy, and learnability.

The study of using computational techniques to analyze and process human languages.

The field of AI that enables systems to interpret and process audio signals such as music or speech.

A field of AI focused on enabling machines to interpret and understand visual inputs from the world.

A process where model performance improves over time during training, ideally reaching stable accuracy.

An optimization algorithm used for classification models and reinforcement learning problems.

The theoretical and practical study of computation, software design, and building intelligent systems.

Raw facts, numbers, or measurements that AI systems use to learn, make predictions, and perform tasks.

A technique to expand existing data by applying small changes, often used in image and text datasets.

The process of fixing or removing errors, inconsistencies, and missing values from raw data.

Reducing the size of data files while preserving the core information, often for storage or transmission.

Discovering patterns and insights in large datasets using AI, statistics, and database systems.

The measure of data's accuracy, completeness, and reliability. High-quality data leads to better AI results.

A structured collection of related data used for training or evaluating AI models.

Organized formats for storing and managing data efficiently in computing environments.

A large centralized data storage system used for analysis and decision-making by AI systems.

Combining AI, analytics, and data to make better, evidence-based decisions.

A mathematical study of how to make optimal decisions under uncertainty, often applied in AI.

A visual model used to make decisions based on branching conditions.

A supervised learning method that creates models based on learned decision rules from data features.

IBM’s chess-playing computer that defeated a world champion, showcasing AI's strength in strategic tasks.

A machine learning technique using deep neural networks to learn patterns and representations from data.

An AI research lab by Google known for AlphaGo and other cutting-edge deep learning breakthroughs.

A model that directly predicts labels from input data without modeling the distribution of features.

AI systems made up of multiple agents working together to solve complex tasks in a decentralized manner.

A technique where random neurons are ignored during training to prevent overfitting.

A field ensuring that AI is developed and used fairly, transparently, and without causing harm.

A program that mimics a human expert's reasoning to solve domain-specific problems using knowledge rules.

The large-scale application of AI across business operations to drive efficiency and intelligence.

The ability to interpret and understand how an AI model arrives at its output.

The ability of an AI system to be expanded or modified without rebuilding its core.

The process of retrieving useful information or features from raw data.

A dense numeric representation of data such as words or images, capturing their contextual meaning.

Unexpected behavior that arises from the complex interactions of AI system components.

A learning model that performs a complete task from raw input to final output in one process.

AI systems designed to be interpretable and transparent in how they make predictions or decisions.

An AI model deployed directly on devices, enabling faster inference without relying on cloud connectivity.

Technology that detects and responds to human emotions based on facial, vocal, or textual cues.

An identifiable object or concept in AI systems, such as a person, place, or organization.

A framework for assessing a company’s sustainability, ethics, and corporate responsibility, often used in AI evaluations.

Processes used to identify and extract key entities like names, locations, and dates from text.

The process of identifying and extracting essential words or phrases from a text.

It is a branch of artificial intelligence that focuses on creating models whose decisions and behaviors can be understood and interpreted by humans. XAI models aim to make AI systems more transparent by providing explanations for how they reach their conclusions. This helps users to trust and understand the reasoning behind AI outputs.

A technology that identifies or verifies a person by comparing their facial features to a database of known faces.

A large pre-trained AI model that can be adapted to many tasks across domains using minimal task-specific data.

The process of moving input data through a neural network to generate predictions.

Identifying key features from raw data that are most useful for model training or decision making.

Adapting a pre-trained AI model to a specific task using additional labeled data.

A learning technique where models learn to perform tasks with very few training examples.

Training AI models across multiple devices without sharing raw data, preserving privacy and efficiency.

An advanced language model capable of generating human-like text and performing various NLP tasks.

A version of GPT known for its powerful text generation and few-shot learning capabilities.

An advanced iteration of GPT offering improved performance in text understanding, generation, and reasoning.

A deep learning architecture with a generator and discriminator that produces highly realistic synthetic data.

An optimization algorithm used to minimize loss functions and train machine learning models efficiently.

A model capable of performing well across various datasets and tasks with minimal fine-tuning.

The process of producing new content such as text, images, or music using generative AI models.

AI models that can create novel content, including text, images, audio, or code, from learned data.

Linking AI models to real-world data to enhance understanding and reliability.

Safety mechanisms designed to limit and control AI behavior and outputs to prevent harmful outcomes.

Problem-solving techniques that use practical methods to produce acceptable solutions quickly.

When an AI model generates false or misleading information not based on the input data.

The process of optimizing configuration settings in AI models to enhance performance and accuracy.

Combining different AI approaches, like symbolic and machine learning, to build more powerful systems.

An approach using simple heuristics and machine learning to solve complex computational problems.

The process where AI identifies objects, people, or features within images.

Improving AI responses by training models on data that includes specific instruction-based prompts.

Using AI tools to boost human decision-making and problem-solving abilities.

The ability to understand how an AI system arrives at its predictions or decisions.

The stage where a trained model makes predictions based on new input data.

Using AI to extract and analyze data from documents automatically.

Refining AI behavior by exposing it to tasks framed with clear instructions.

Enhancing human cognition through AI tools to analyze, process, and visualize information better.

An algorithm for efficient probabilistic inference in graphical models like Bayesian networks.

The process where models learn from massive data and produce new insights or answers.

An approach where AI learns to classify with only K labeled examples per class.

The process of encoding expert knowledge into systems that simulate human decision-making.

A structured network of entities and their relationships used to organize and infer knowledge.

A framework for how information is organized and related within a system.

AI systems that reason using stored knowledge bases, rules, and logic rather than learning from data.

AI models trained on massive text datasets to understand and generate human-like language.

The delay between a user’s input and the AI system’s response, often measured in milliseconds.

A supervised learning method that models relationships between input variables and outputs using a straight line.

A compressed space in neural networks that holds learned patterns and abstract features from input data.

The tools and processes used to manage, optimize, and monitor AI language model performance and outputs.

Text-based content used to train and evaluate language models for better language understanding.

A formula that measures how far an AI model’s predictions are from actual results.

A type of AI that uses temporary memory of past data to improve current decisions.

The base or dictionary form of a word used in NLP after lemmatization, e.g., 'run' from 'running'.

A collection of words and meanings used by AI models to understand and process language.

A statistical method for binary classification, often used to predict categories like true/false or spam/not spam.

An AI method where systems learn patterns from data to make predictions or decisions without explicit programming.

High-level prompts or context settings that guide an AI model’s behavior during training or inference.

Data that describes other data, providing context like source, format, or time of creation.

A trained algorithm used by AI to make predictions or analyze data based on learned patterns.

When an AI model’s accuracy declines over time due to changing data or conditions.

Internal variables in a model that are learned during training and help define how the model behaves.

Breaking down language or problems into smaller parts for easier understanding and processing by AI.

AI models trained to understand different types of data like images, audio, and text together.

A technique where AI is trained to handle multiple similar tasks through custom prompt structures.

A learning method where the AI model is trained on a small number (N) of examples per class.

The presence of multiple meanings in language that can confuse AI interpretation.

A technology that creates human-like text or speech from structured data.

AI techniques that help computers understand, interpret, and generate human language.

The subfield of NLP that focuses on interpreting meaning, context, and emotions in human language.

A method of building applications without programming, enabling anyone to create software visually.

An AI system inspired by the human brain, composed of layers of interconnected nodes (neurons).

A technique to reconstruct 3D scenes from 2D images using neural networks.

A way to ask questions using everyday language instead of code or structured syntax.

A field combining AI and linguistics to build systems that understand and use human language.

A research company focused on creating safe and beneficial AI, known for ChatGPT and GPT-4.

Improving a model’s performance by minimizing prediction errors or maximizing accuracy.

When an AI model memorizes training data too closely and performs poorly on new data.

A formula that a model optimizes during training to achieve the best possible performance.

A system that organizes information into categories and defines relationships among them for AI understanding.

PEFT is a method to improve large AI models by modifying only a few critical components, reducing time and energy usage.

The first stage of AI model training, using general data before fine-tuning on task-specific data.

The practice of crafting optimal prompts to guide AI models for accurate and useful outputs.

A model that predicts outcomes based on probability rather than certainty, accounting for multiple possibilities.

Internal values of a model adjusted during training to improve prediction accuracy.

Training machines to identify trends or regularities in data for classification or prediction.

Using historical data and models to forecast future trends and behaviors.

Analytics that recommends actions by analyzing options, outcomes, and resources.

An instruction or query given to an AI system to generate a response.

Analyzing and breaking down text into grammatical components to understand sentence structure.

Labeling each word in a sentence with its grammatical role like noun, verb, or adjective.

The process of human correction of machine-translated content to improve quality.

Add-on tools that expand the capabilities of AI systems, enabling tasks like browsing or calculation.

Refining or filtering AI-generated outputs after prediction for better clarity or usefulness.

Cleaning and organizing raw data into usable form before feeding it into AI models.

Learning general skills from large datasets before fine-tuning a model on specific tasks.

The percentage of correct positive predictions out of all positive predictions made by a model.

A model that has already learned general knowledge and can be adapted for specific tasks.

Using a series of prompts to guide AI through step-by-step reasoning and refine outputs.

Using quantum physics principles to perform computations faster than classical computers, especially in AI applications.

A machine learning technique that builds multiple decision trees to improve prediction accuracy.

The percentage of actual positives correctly identified by the model out of all possible positives.

Neural networks that use loops to retain previous information, often used for sequential data like speech or text.

A training method where AI learns by trial and error through rewards and punishments.

Enhancing reinforcement learning by incorporating human preferences into model training.

In NLP, refers to the connection between words or phrases, such as familial or semantic relationships.

Combining large language models with external sources of information to generate more factual outputs.

A metric that evaluates the success or value of investing in AI technologies.

A translation method using fixed grammatical rules to convert one language into another.

The integration of AI in physical machines to enhance automation, productivity, and efficiency.

A method to prevent overfitting by simplifying the model and adding penalty terms to the loss function.

The ability of AI to make decisions or solve problems through logic and inference.

A technique that refines AI responses through a series of layered prompts, each building on the last.

A supervised learning is a type of machine learning where the model is trained on labeled data, meaning the input data is paired with the correct output. With this approach the AI model learns to predict the output from new inputs.

A supervised learning is a type of machine learning where the model is trained on labeled data, meaning the input data is paired with the correct output. With this approach the AI model learns to predict the output from new inputs.

Speech-to-text is a technology that converts spoken language into written text by analyzing and interpreting audio recordings.

Symbolic artificial intelligence is a branch of AI that uses precise, human-readable symbols and rules to represent knowledge and solve problems, often through logic and reasoning.

Stable Diffusion is an advanced AI method used to generate images or other media by continuously refining noisy data in prompts into a clearer form. This AI technology is commonly used in generative AI models.

The term Sentiment Analytics refers to the process of using AI to analyze and determine the emotional tone (positive, negative, neutral) behind a piece of text. This mechanism gives insights of a textual content and distinguishes whether the content is a product review or social media post.

It is the capability to adjust or control the behavior or output of an AI model, often in a guided or targeted manner.

The set of rules that govern the structure of sentences or expressions in a language determines how words and symbols are organized.

This term refers to a component in neural networks, especially in transformers, that allows the AI model to consider the importance of different words in a sentence when making predictions. While making predictions, it focuses on relevant parts of the input.

The term Strong AI refers to a hypothetical form of artificial intelligence that possesses general intelligence. It means that the Strong AI can understand, learn, and apply knowledge across a wide range of tasks, much like a human.

It is a search technique that understands the meaning behind words and phrases to provide more accurate and relevant search results beyond simple keyword matching.

The term Structured Data refers to the data that is organized in a predefined format, such as rows and columns in a spreadsheet or database, making it easy to store, search, and analyze.

It is the process of automatically condensing a large body of text into a shorter version while retaining the most important information.

Symbolic AI is a type of artificial intelligence that uses symbols and logical rules to represent knowledge and perform reasoning, typically involving human-understandable concepts.

The term Symbolic Methodology refers to an approach in AI where problems are solved using symbolic representations (like logic, rules, and symbols) rather than through statistical methods.

It is the use of AI to analyze speech recordings, extracting insights such as emotional tone, key themes, or the speaker’s intent from the audio.

Semantics is the study of meaning in language, focusing on how words, phrases, and sentences convey meaning.

A form of data that does not follow a strict structure but has some organizational properties, such as tags or markers, which makes it easier to analyze than unstructured data.

This terminology refers to the technology that converts spoken language into text by identifying and transcribing the words spoken in an audio recording.

Structured data is data organized into a clear, predefined format, often in databases or tables, making it easy to search, analyze, and manage.

Sentiment analysis is a process of using AI to identify and classify the emotional tone of a piece of text. Such kind of analysis is used to determine whether a review is positive, negative, or neutral.

Supervised Learning is a type of machine learning where the AI model is trained on labeled data, meaning the input data comes with the correct output, and the model learns to predict the output from new inputs.

A term used to describe large language models that generate text by mimicking patterns in their training data without true understanding, often producing fluent but contextually shallow responses.

Similarity is the measure of how alike two objects are, often used to compare features or data points.

Semantic Networks, or SN, refer to a knowledge representation structure that shows concepts and their relationships. These networks are often used to represent meaning and associations in AI systems.

This term refers to collections of texts or data sets that are specifically designed or compiled for a particular subject area, domain, or language use. These are often used in training AI models for specialized tasks.

Singularity is a hypothetical future point when artificial intelligence surpasses human intelligence, leading to rapid technological advancements that could be unpredictable or uncontrollable.

This term refers to a standardized framework for organizing and sharing knowledge using concepts, labels, and relationships. It is often used to structure information in digital libraries and the Semantic Web.

Transformer is a type of neural network architecture used in AI that processes and understands sequences of data. They focus on the most relevant parts of the text input or prompt.

This term refers to an open-source machine learning framework created by Google that helps developers build and train AI models, particularly for deep learning tasks.

The token is a piece of a larger text, such as a word, punctuation mark, or symbol, used in natural language processing to break down and analyze text data.

The Turning Test, also known as the Turing Test, was proposed by Alan Turing to assess a machine’s capability to exhibit intelligent behavior distinguishable from that of a human during a conversation.

Tagging refers to the process of labeling or annotating data, such as words in a text, with categories or identifiers (e.g., parts of speech or named entities) to make it easier for AI models to analyze.

This term refers to parameters or settings in a machine learning model that can be adjusted or fine-tuned to improve the model’s performance.

It is a way of representing knowledge where information is broken down into three parts: a subject (who/what is doing something), an action (what they are doing), and an object (what is being acted upon).

This term refers to a visualization tool that displays hierarchical data as a set of nested rectangles, where the size and color of each rectangle represent the data’s importance or value.

TPU is a specialized hardware chip designed by Google to accelerate machine learning tasks, particularly for deep learning models, by efficiently processing tensors.

It refers to a technology that converts written text into spoken words by synthesizing audio that sounds like human speech.

The labeled data used to train a machine learning model, helping the model learn patterns and relationships to make predictions on new, refers to Training Data.

Tokenization is a process of breaking down a text into smaller pieces, such as words, phrases, or symbols, to make it easier for a machine to understand and analyze.

It is a technique in machine learning in which an AI model trained on one task is reused or adapted to perform a different but related task. This type of learning saves time and improves the overall performance of an AI model.

The term Training Set refers to a subset of labeled data used specifically to train a machine learning model. It helps the AI models to learn patterns and relationships before being tested on new data.

It is a type of machine learning in which an AI model is trained on data without labeled answers. The model tries to find hidden patterns or groupings within the data on its own.

This term refers to an AI model’s poor performance. It happens when a machine learning model is too simple and fails to learn the patterns in the data, leading to poor performance on both the training data and new data.

It is a type of data that doesn’t have a predefined format or organization, like text, images, or videos, making it harder to analyze compared to structured data, like tables.

This term refers to a software program that uses AI to perform tasks or services for a user, such as answering questions, setting reminders, or controlling smart devices, often through voice or text interactions (e.g., Siri, Alexa).

Voice Processing refers to the technology used to analyze, interpret, and manipulate voice signals, such as converting spoken language into text or recognizing speaker identity.

It is a separate set of labeled data used during training to evaluate and adjust a machine learning model’s performance. This data ensures that the model generalizes well to new, unseen data.

It is a separate set of labeled data used during training to evaluate and adjust a machine learning model’s performance. This data ensures that the model generalizes well to new, unseen data.

A VPU is a specialized processor designed to accelerate computer vision tasks, such as recognizing objects in images or videos, by efficiently handling the data required for these tasks.

Weak AI, also known as narrow AI, refers to artificial intelligence that is designed and trained to perform a specific task or a limited set of tasks. Unlike strong AI, weak AI does not possess general intelligence or understanding beyond its designated function.

In AI, yield management systems use algorithms and machine learning models to analyze large amounts of data (such as demand, booking patterns, and competition) to adjust prices dynamically in real-time. AI-powered yield management systems help businesses make smarter pricing decisions, predict customer behavior, and allocate resources more effectively to maximize profitability.

This term refers to a machine learning technique where a model is able to recognize and classify objects or categories that it has never been explicitly trained on by using knowledge transferred from related tasks or features.