The realm of Artificial Intelligence (AI) is a dynamic field marked by rapid advancements and innovations. As AI continues to permeate various sectors, understanding its foundational concepts and terminology is crucial for professionals, enthusiasts, and the curious alike. This comprehensive A-to-Z glossary demystifies essential AI terms, offering a window into the fascinating world of intelligent machines.
A:
Algorithms
Definition: An algorithm in AI is a set of rules or a sequence of instructions designed to perform a specific task or solve a particular problem. Algorithms are the heart of AI systems, enabling them to process data, learn from it, and make decisions or predictions.
Artificial Intelligence (AI)
Definition: Artificial Intelligence is the branch of computer science that focuses on creating machines capable of mimicking human intelligence. This includes learning, reasoning, problem-solving, perception, and language understanding. AI can be categorised into two types: Narrow AI, which is designed for specific tasks, and General AI, which has broader, human-like capabilities.
Artificial Neural Networks (ANN)
Definition: Artificial Neural Networks are computing systems inspired by the biological neural networks that constitute animal brains. An ANN is composed of interconnected units or nodes called artificial neurons, which process information using a connectionist approach to computation. ANNs are fundamental to deep learning and are used for a variety of tasks, including image and speech recognition.
Augmented Reality (AR)
Definition: Augmented Reality is an interactive experience where real-world environments are enhanced by computer-generated perceptual information. In the context of AI, AR can be powered by intelligent algorithms to provide users with enriched, context-aware experiences, overlaying digital content onto the physical world.
Autonomous Vehicles
Definition: Autonomous vehicles, also known as self-driving cars, are vehicles equipped with AI systems that can navigate and operate without human intervention. These vehicles use a combination of sensors, cameras, radar, and AI algorithms to perceive their surroundings, make decisions, and navigate safely.
Automation
Definition: Automation refers to the use of technology to perform tasks with minimal human assistance. In AI, automation often involves using algorithms to process data, make decisions, or carry out actions that would typically require human intelligence, such as natural language processing or image recognition.
AGI (Artificial General Intelligence)
Definition: Artificial General Intelligence is a theoretical form of AI that has the ability to understand, learn, and apply its intelligence to solve any problem with the same competence as a human. AGI encompasses the understanding and reasoning across a wide range of domains, a milestone AI researchers aim to achieve.
B:
Backpropagation: A fundamental algorithm used in training artificial neural networks. It efficiently computes the gradient of the loss function with respect to the weights of the network, allowing for the adjustment of weights to minimise error. Backpropagation is key to the learning process in deep learning models.
Bayesian Networks: A type of probabilistic graphical model that uses Bayesian inference for probability computations. Bayesian networks represent a set of variables and their conditional dependencies via a directed acyclic graph (DAG), enabling complex reasoning under uncertainty.
Bias: In machine learning, bias is a systematic error in predictions, due to assumptions in the learning algorithm. High bias can cause an algorithm to miss relevant relations between features and target outputs (underfitting). Bias can also refer to unfair and prejudiced outcomes produced by an AI system, often reflecting existing societal biases.
Big Data: Extremely large data sets that may be analysed computationally to reveal patterns, trends, and associations, especially relating to human behaviour and interactions. Big data is essential for training more accurate and sophisticated AI models.
Binary Classification: A type of classification task where the output is restricted to two classes. For example, an email classification system may categorise emails into 'spam' and 'not spam'.
Bioinformatics: The application of computational technology to handle biological and genetic data. AI in bioinformatics is used for sequences analyses, gene expression, and prediction of gene functions, among other applications.
Blockchain: A distributed database or ledger that is shared among the nodes of a computer network. While primarily associated with cryptocurrencies, blockchain technology is also being explored for AI applications, particularly in ensuring data integrity and security.
Bot: Short for robot, it refers to an AI system or software application designed to perform automated tasks, such as answering questions, performing online searches, or controlling smart home devices. Bots range from simple scripted services to advanced AI-driven assistants.
Business Intelligence (BI): The strategies and technologies used by enterprises for data analysis and business information management. AI enhances BI tools with predictive analytics, data mining, and machine learning to transform raw data into insightful business actions.
Byte: A unit of digital information in computing and telecommunications that commonly consists of eight bits. Understanding data volume units like bytes is crucial when dealing with the large amounts of data processed and generated by AI systems.
C:
Chatbot: A software application used to conduct an online chat conversation via text or text-to-speech, in lieu of providing direct contact with a live human agent. Designed to convincingly simulate the way a human would behave as a conversational partner, chatbots are typically used in customer service or information acquisition scenarios.
Clustering: A method of unsupervised learning where AI systems group a set of objects in such a way that objects in the same group (called a cluster) are more similar to each other than to those in other groups. It's widely used in data mining, pattern recognition, and image analysis.
Cognitive Computing: A subset of AI that attempts to mimic human thought processes in a computerised model. It involves self-learning systems that use data mining, pattern recognition, natural language processing, and human senses emulation to mimic the way the human brain works.
Computer Vision: A field of AI that trains computers to interpret and understand the visual world. Using digital images from cameras and videos and deep learning models, computers can accurately identify and classify objects, and then react to what they “see.”
Convolutional Neural Network (CNN): A class of deep neural networks, most commonly applied to analysing visual imagery. CNNs are inspired by biological processes and are variations of multilayer perceptrons designed to use minimal amounts of preprocessing.
Cross-validation: A statistical method used in machine learning to evaluate the performance of a model. It involves partitioning the data into subsets, training the model on one subset, and validating it on another to check its predictive performance.
Crowdsourcing: The practice of obtaining input or information for a task or project by enlisting the services of a large number of people, either paid or unpaid, typically via the Internet. In AI, crowdsourcing can be used to gather datasets for training models or to refine AI algorithms through human feedback.
Cybernetics: An interdisciplinary approach for exploring regulatory systems, their structures, constraints, and possibilities. In AI, cybernetics has influenced the development of systems that can simulate decision-making and problem-solving processes.
CUDA (Compute Unified Device Architecture): A parallel computing platform and application programming interface (API) model created by NVIDIA. It allows software developers and software engineers to use a CUDA-enabled graphics processing unit (GPU) for general purpose processing – an approach known as GPGPU (General-Purpose computing on Graphics Processing Units). CUDA is highly relevant in deep learning for accelerating neural network computations.
D:
Data Mining: The process of discovering patterns and knowledge from large amounts of data. The data sources can include databases, data warehouses, the internet, and other data repositories. Data mining involves statistical analysis, machine learning, and database systems to uncover insights.
Data Science: An interdisciplinary field that uses scientific methods, processes, algorithms, and systems to extract knowledge and insights from structured and unstructured data. Data science is closely related to data mining and big data.
Deep Learning: A subset of machine learning based on artificial neural networks with representation learning. Deep learning architectures, such as deep neural networks, enable machines to process data in complex layers, leading to significant advancements in areas like image and speech recognition.
Decision Tree: A decision support tool that uses a tree-like model of decisions and their possible consequences, including chance event outcomes, resource costs, and utility. It's a way to display an algorithm that contains only conditional control statements.
Dimensionality Reduction: The process of reducing the number of random variables under consideration, by obtaining a set of principal variables. Techniques like Principal Component Analysis (PCA) are used in machine learning to simplify models and speed up computation.
Discriminative Model: A type of model in machine learning that models the decision boundary between the classes of the data. It learns the conditional probability distribution P(Y|X) – the probability of Y given X.
Distributed Computing: A field of computer science that studies distributed systems. In the context of AI, it refers to the distribution of processes and data across multiple machines and the coordination of computations through message passing or shared memory.
Domain Knowledge: Specific knowledge or expertise in a particular domain that an AI system can be trained on or use to enhance its performance. Incorporating domain knowledge into AI models can significantly improve their accuracy and effectiveness.
Dynamic Programming: A method for solving complex problems by breaking them down into simpler subproblems. It is applicable to problems exhibiting the properties of overlapping subproblems and optimal substructure. In AI, it's used in various algorithms, including those for reinforcement learning.
DeepFake: A technique for human image synthesis based on powerful deep learning algorithms. DeepFake technology can create convincing but entirely fictional video and audio recordings, posing significant challenges for security, privacy, and information integrity.
E:
Ensemble Learning: A machine learning paradigm where multiple models (often of varying types) are trained to solve the same problem and then aggregated to obtain better results. The rationale behind ensemble methods is that many weak models working together can outperform a single very accurate model by reducing overfitting.
Evolutionary Algorithms: A subset of artificial intelligence that uses mechanisms inspired by biological evolution, such as reproduction, mutation, recombination, and selection. These algorithms are used to generate high-quality solutions to optimisation and search problems by relying on bio-inspired operators such as mutation, crossover, and selection.
Expert System: An AI program that mimics the decision-making ability of a human expert. Expert systems are designed to solve complex problems by reasoning through bodies of knowledge, represented mainly as if-then rules rather than conventional procedural code.
Explainable AI (XAI): An emerging field in machine learning that addresses how black box decisions of AI systems are made. XAI aims to make AI models more transparent and understandable to humans, allowing for greater trust and reliability in AI applications, especially in critical areas like healthcare, finance, and law.
Embeddings: In the context of machine learning, embeddings are low-dimensional learned representations of data in a high-dimensional space. Text or words are commonly converted into embeddings so that they can be processed in neural network models, enabling the machine to understand the semantic relationships between words.
Ethics in AI: Refers to the moral principles and techniques employed to ensure that AI technology is designed, developed, and used in a way that is fair, accountable, and devoid of bias. Ethics in AI also encompasses considerations of privacy, transparency, equality, and security.
Early Stopping: A form of regularisation used to avoid overfitting when training a learner with an iterative method, such as gradient descent. The method entails stopping the training process if the performance on a validation dataset decreases or ceases to improve significantly.
Entity Recognition: Also known as Named Entity Recognition (NER), it is a subtask of information extraction that seeks to locate and classify named entities mentioned in unstructured text into predefined categories such as the names of persons, organisations, locations, expressions of times, quantities, monetary values, percentages, etc.
Echo State Network (ESN): A type of recurrent neural network with a sparsely connected hidden layer. The connectivity and weights of hidden units are fixed and only the weights from the hidden units to the output units are learned. ESNs are particularly suited for temporal data processing.
Edge Computing: A distributed computing paradigm that brings computation and data storage closer to the location where it is needed, to improve response times and save bandwidth. In AI, edge computing facilitates the deployment of AI models directly on devices (like smartphones and IoT devices), enabling faster and more efficient processing.
F:
Feature Engineering: The process of using domain knowledge to extract features (characteristics, properties, attributes) from raw data. These features can be used to improve the performance of machine learning algorithms. Feature engineering is crucial because the right features can simplify the learning process and enhance model accuracy.
Fuzzy Logic: A form of many-valued logic or probabilistic logic; it deals with reasoning that is approximate rather than fixed and exact. Fuzzy logic variables may have a truth value that ranges in degree between 0 and 1. Fuzzy logic is used in some AI systems to handle situations where the information is uncertain, ambiguous, or incomplete.
Federated Learning: A machine learning approach where the training algorithm is distributed across many devices or servers holding local data samples and does not exchange them. Instead, the updated gradients or parameters are shared for aggregation, keeping the data localised and improving privacy and security.
Forward Chaining: A method in rule-based expert systems that starts with the available data and uses inference rules to extract more data (in a forward manner) until a goal is reached. It is a bottom-up approach used commonly in logical reasoning and AI.
False Positive & False Negative: In the context of AI and machine learning, a false positive is an error in data reporting in which a test result improperly indicates the presence of a condition (such as a disease when it is not present), while a false negative is an error in which a test result improperly indicates no presence of a condition (when it is present).
Frame Problem: In artificial intelligence, the frame problem describes the challenge of specifying what is assumed to remain unchanged in a given situation, after certain changes have occurred. This problem arises because, in a logical system, an action does not entail non-effects, making it hard to infer what remains the same.
Facial Recognition: A technology capable of identifying or verifying a person from a digital image or a video frame from a video source. Facial recognition systems have been deployed in various applications, from security systems to marketing and personal device authentication.
Feedback Loop: A process in which the outputs of a system are circled back and used as inputs. In the context of AI, feedback loops can either enhance the learning process through positive feedback or correct errors and refine models through negative feedback.
Fully Connected Layer: In neural networks, a fully connected layer is a layer where each neuron is connected to every neuron in the previous layer. These layers are often placed near the end of neural networks and are used to combine learned features from previous layers for the task of classification or regression.
Fine-tuning: In machine learning, fine-tuning involves making small adjustments to a model that has already been trained on a similar task to adapt it to a specific task. It's particularly used in transfer learning, where a model developed for one task is reused as the starting point for a model on a second task.
G:
Generative Adversarial Networks (GANs): A class of machine learning frameworks designed by Ian Goodfellow and his colleagues in 2014. Two neural networks contest with each other in a game (thus, "adversarial"). One network generates candidates (generative) and the other evaluates them (discriminator). GANs are used to generate realistic images, videos, and voice outputs.
Gradient Descent: An optimisation algorithm used to minimise some function by iteratively moving in the direction of steepest descent as defined by the negative of the gradient. It is used in machine learning and deep learning for training models.
Graph Neural Networks (GNNs): A type of neural network that directly operates on the graph structure. GNNs take graph nodes, edges, and other structural features into account in processing the data, making them suitable for tasks like social network analysis, recommendation systems, and drug design.
Genetic Algorithms: An optimisation technique based on the principles of genetics and natural selection. It is used in AI to solve optimisation and search problems by iteratively selecting, breeding, and mutating populations of solutions.
Ground Truth: The accuracy of training set's classification for supervised learning in the context of machine learning and AI. It is used as a benchmark to measure the performance of AI models. Ground truth represents the reality to be modelled or predicted.
Greedy Algorithm: A simple, intuitive algorithm that is used in optimisation problems. The algorithm makes the locally optimal choice at each stage with the hope of finding the global optimum. In AI, greedy algorithms are used for decision-making processes.
Gaussian Process: A probabilistic model wherein observations occur in a continuous domain, e.g., time or space. In machine learning, Gaussian processes are used for regression and classification tasks, particularly in models that need to predict uncertain outcomes.
GPT (Generative Pre-trained Transformer): A type of artificial intelligence model used for natural language processing tasks. It is pre-trained on a large corpus of text and then fine-tuned for specific tasks like translation, question-answering, and text generation. GPT models are known for their ability to generate coherent and contextually relevant text based on prompts.
Game Theory: A mathematical framework designed for analysing situations among competing players and strategising under defined rules and outcomes. In AI, game theory is applied to model and solve problems where agents interact with each other, making decisions that affect the overall outcome.
General AI: Also known as Artificial General Intelligence (AGI), it refers to machines that possess the ability to understand, learn, and apply intelligence to solve any problem, similarly to a human. Unlike narrow AI, AGI can generalise learning across domains. This level of AI sophistication remains a theoretical concept as of now.
H:
Heuristic: In the context of AI, a heuristic is a problem-solving approach that employs a practical method not guaranteed to be optimal but sufficient for reaching an immediate, short-term goal or approximation. Heuristics play a crucial role in decision-making processes, especially in complex problems where finding an optimal solution is impractical due to time constraints.
Hybrid Model: In artificial intelligence, a hybrid model combines characteristics of both rule-based and statistical AI systems. These models leverage the structured reasoning of rule-based systems along with the adaptive learning capabilities of statistical models to improve performance and decision-making accuracy.
Hyperparameters: These are the parameters whose values are set before the learning process begins, unlike other parameters which are derived via training. Hyperparameters define higher-level concepts about the model such as its complexity or how fast it should learn, and they can significantly impact the performance of machine learning algorithms.
Hyperplane: In machine learning, particularly in support vector machines, a hyperplane is a decision boundary that helps to classify data points. In higher-dimensional spaces, a hyperplane is a flat affine subspace whose dimension is one less than that of its ambient space, used to separate different classes of data.
Homomorphic Encryption: A form of encryption that allows computation on ciphertexts, generating an encrypted result which, when decrypted, matches the result of operations performed on the plaintext. This is an important area of research in AI, enabling users to perform analyses on encrypted data without compromising privacy.
Human-in-the-Loop (HITL): A model of interaction where a human is directly involved in the training, tuning, or testing of the AI algorithms. HITL approaches are particularly valuable in scenarios where AI systems must adapt to new or unexpected situations that were not covered in their initial training data.
Haptic Technology: Also known as kinesthetic communication or 3D touch, refers to any technology that can create an experience of touch by applying forces, vibrations, or motions to the user. In AI, haptic feedback is used to enhance the interactivity of virtual and augmented reality systems, providing users with tangible feedback in response to actions or events within a simulation.
Hebbian Learning: Named after psychologist Donald Hebb, this is a theory that proposes an explanation for the adaptation of neurons in the brain during the learning process, described as "cells that fire together, wire together." In AI, Hebbian learning principles are used in neural network models to simulate how neurons strengthen with use over time.
Heterogeneous Computing: The use of systems that incorporate multiple types of processors or cores, which may include CPUs, GPUs, and specialised accelerators, to perform computational tasks. In AI, heterogeneous computing architectures are often employed to speed up the training and inference phases of machine learning models by leveraging the strengths of different processing units.
Hidden Markov Model (HMM): A statistical model used to represent systems that are assumed to be Markov processes with unobserved (hidden) states. HMMs are used in various AI applications, including speech recognition, handwriting recognition, and bioinformatics, for analysing temporal sequences.
I:
Intelligent Agents: Entities that perceive their environment through sensors and act upon that environment with effectors. Intelligent agents can be simple, like a thermostat controlling a room's temperature, or complex, such as autonomous robots or virtual personal assistants.
Image Recognition: A computer vision technique that allows machines to interpret and categorise what they "see" in images or videos. Image recognition algorithms use patterns to differentiate objects within an image, applying to various applications from social media photo tagging to medical diagnosis.
Incremental Learning: A machine learning approach where the model is continuously updated with new data without needing to retrain from scratch. This allows AI systems to adapt to new information over time and is particularly useful for applications where data arrives in streams or the environment changes.
Inference Engine: The component of an expert system that applies logical rules to the knowledge base to deduce new information or make decisions. Inference engines are crucial for systems requiring decision-making capabilities based on complex, rule-based logic.
IoT (Internet of Things): A network of physical objects (things) embedded with sensors, software, and other technologies for the purpose of connecting and exchanging data with other devices and systems over the internet. AI plays a significant role in IoT by enabling devices to analyse and act on the data they collect, enhancing efficiency and enabling new services.
Inductive Reasoning: A logical process in which multiple premises, all believed true or found true most of the time, are combined to obtain a specific conclusion. In AI, inductive reasoning is used to form generalisations based on observations and examples.
Information Retrieval: The process of obtaining relevant information from a collection of resources in response to a query. In AI, information retrieval systems are powered by algorithms that can sift through vast amounts of data to find meaningful and relevant results, such as search engines.
Instance-based Learning: A family of learning algorithms that compare new problem instances with instances seen in training, which were stored in memory. k-Nearest Neighbors (k-NN) is a well-known example where the algorithm classifies new cases based on their similarity to instances seen during training.
Integrated Development Environment (IDE): A software application that provides comprehensive facilities to computer programmers for software development. An IDE typically includes a code editor, compiler or interpreter, and debugger, all accessible from a single graphical user interface (GUI). For AI development, IDEs often include tools and libraries specifically designed for machine learning and data science tasks.
Imitation Learning: A technique where AI models learn to perform tasks by mimicking human actions. Also known as learning from demonstration, imitation learning is used in scenarios where defining explicit rules is difficult, allowing robots or virtual agents to learn complex behaviours directly from human examples.
J:
Joint Probability Distribution: In statistics and AI, the joint probability distribution for a set of variables is a probability distribution that specifies the likelihood of different combinations of those variables' values occurring simultaneously. Understanding joint distributions is crucial for many AI applications, including probabilistic graphical models and Bayesian networks.
Jupyter Notebook: An open-source web application that allows you to create and share documents that contain live code, equations, visualisations, and narrative text. Jupyter Notebooks are widely used in data science and AI for data cleaning and transformation, numerical simulation, statistical modelling, machine learning, and much more.
JavaScript Object Notation (JSON): A lightweight data-interchange format that is easy for humans to read and write and easy for machines to parse and generate. In AI, JSON is often used for configuring machine learning models, describing datasets, and exchanging data between servers and web applications.
Jacobian Matrix: In mathematics and AI, the Jacobian matrix consists of all first-order partial derivatives of a vector-valued function. In the context of neural networks and deep learning, the Jacobian matrix can be used to understand how changes in the input of a network affect changes in the output, which is important for optimization and understanding model behaviour.
Java AI Libraries: Various libraries in Java are designed to support AI development, offering tools and functions for machine learning, neural networks, natural language processing, and more. Some popular Java AI libraries include Deeplearning4j, Weka, and MOA (Massive Online Analysis).
Jumpstart: In AI development, jumpstarting refers to the practice of using pre-trained models or existing codebases as a starting point for a new project. This approach can significantly speed up the development process by leveraging work that has already been done, allowing developers to focus on customising and extending the model for their specific needs.
Just-In-Time Compilation (JIT): A compilation approach used in runtime environments, such as Java Virtual Machine (JVM) or .NET Framework, which compiles code into machine language at the moment it is needed during execution, rather than prior to execution. JIT compilation can improve the performance of AI applications by optimising code execution dynamically.
Jensen-Shannon Divergence: A method of measuring the similarity between two probability distributions. It is symmetric and always has a finite value, making it useful for comparing distributions in various AI tasks, including generative modelling and document similarity analyses.
Job Scheduling: In the context of AI and computing, job scheduling is the process of assigning system resources to perform different tasks according to a plan. Efficient job scheduling is crucial in environments where AI models are trained and deployed, ensuring that computational resources are used optimally.
Jargon Tokenisation: In natural language processing (NLP), tokenisation is the process of breaking down text into smaller units called tokens, which could be words or phrases. Jargon tokenization specifically refers to the identification and handling of domain-specific terminology, which is important for applications like medical or legal document analysis, where understanding specialised jargon is critical.
K:
Knowledge Base: In artificial intelligence, a knowledge base is a centralised repository for information: a public library, a database of related information about a particular subject. It's designed to allow computers to process the knowledge stored in it and use it to answer complex questions or solve specific problems through reasoning.
Knowledge Representation: A field in AI dedicated to representing information about the world in a form that a computer system can utilise to solve complex tasks such as diagnosing a medical condition or communicating with humans in natural language. It involves the abstraction and encoding of complex real-world data into understandable formats for machines.
K-nearest Neighbours (k-NN): A simple, versatile, and easy-to-implement supervised machine learning algorithm used for classification and regression. It classifies or predicts the value of a data point based on the majority vote or average of its 'k' nearest neighbours in the feature space.
K-means Clustering: An unsupervised learning algorithm that aims to partition 'n' observations into 'k' clusters in which each observation belongs to the cluster with the nearest mean. It's widely used for statistical data analysis in various fields, including market segmentation, pattern recognition, and image analysis.
Knowledge Engineering: The process of creating rules that apply logic to knowledge base content in an expert system, enabling it to make informed decisions. Knowledge engineering involves gathering knowledge from experts and structuring it in a way that machines can interpret and use it effectively.
Keras: An open-source neural network library written in Python, designed to enable fast experimentation with deep neural networks. It acts as an interface for the TensorFlow library and simplifies many aspects of creating and compiling deep learning models.
Kernel Methods: A class of algorithms for pattern analysis, whose best known member is the support vector machine (SVM). They work by mapping input data into high-dimensional feature spaces, making it easier to perform linear separation or regression on complex datasets.
Keyframe Extraction: In the context of video processing and computer vision, keyframe extraction involves selecting a subset of frames from a video that effectively summarises its entire content. This technique is useful in various AI applications, including video summarisation and indexing.
Knowledge Discovery in Databases (KDD): The process of discovering useful knowledge from a collection of data. This interdisciplinary subfield of computer science involves statistics, database systems, machine learning, and artificial intelligence, among others, to extract patterns and knowledge from large datasets.
Kurtosis: A statistical measure used to describe the distribution of observed data around the mean. In AI and machine learning, understanding the kurtosis of the data can help in preprocessing steps and in choosing the right models or algorithms for accurate predictions.
Knowledge Graph: A knowledge base that uses a graph-structured data model or topology to integrate data. Knowledge graphs are used to store interlinked descriptions of entities – objects, events, or concepts – with free-form semantics. They are particularly useful in semantic searches, recommendation systems, and AI applications that require a rich understanding of relationships and properties within data.
L:
Learning Rate: In machine learning, the learning rate is a parameter that determines the step size at each iteration while moving toward a minimum of a loss function. It affects how quickly a model can converge to a local minimum and how well it generalises to new data.
Latent Variable: A variable that is not directly observed but is inferred or estimated from observed variables within a mathematical model. In AI, latent variables are often used in probabilistic models and latent semantic analysis to capture underlying structures in data.
Logistic Regression: A statistical method for analysing a dataset in which there are one or more independent variables that determine an outcome. The outcome is measured with a dichotomous variable (where there are only two possible outcomes). In machine learning, logistic regression is used for binary classification problems.
Long Short-Term Memory (LSTM): A special kind of recurrent neural network (RNN) capable of learning long-term dependencies. LSTMs are particularly useful for processing sequences of data such as time series or natural language, making them effective for tasks like speech recognition, language modelling, and text generation.
Loss Function: A method for evaluating how well a specific algorithm models the given data. If predictions deviate from actual results, loss functions provide a measure of the error. Minimising this error during training is a primary goal in machine learning.
Label: In supervised learning, a label is the output or target variable that a model predicts. It is the answer or result for a given observation. Labels are used during the training phase to teach the model the correct output for a given input.
Language Model: A statistical machine learning model that is trained to predict the probability of a sequence of words. Language models are fundamental to various natural language processing tasks such as speech recognition, text generation, and machine translation.
Labelled Data: Data that has been tagged with one or more labels identifying certain properties or classifications. Labelled data is used in supervised learning to train models, where the model learns from the input data and the corresponding labels.
Latent Semantic Analysis (LSA): A technique in natural language processing of analysing relationships between a set of documents and the terms they contain by producing a set of concepts related to the documents and terms. LSA is used to extract the hidden (latent) relationships between words in large text corpora.
Linear Regression: A linear approach to modelling the relationship between a dependent variable and one or more independent variables. In machine learning, linear regression is used for predictive modelling, forecasting numerical values based on input features.
Leakage: In machine learning, leakage occurs when information from outside the training dataset is used to create the model. This can lead to overly optimistic performance measures during training and testing and poor performance on new, unseen data.
Layer (Neural Networks): A collection of neurons within a neural network. Layers are structured in a hierarchical manner, with each layer performing specific transformations on its inputs. Neural networks typically consist of input, hidden, and output layers.
M:
Machine Learning (ML): A subset of artificial intelligence focused on building systems that learn from data. Unlike traditional algorithms, machine learning models adjust their actions or predictions based on the patterns they detect in the data, improving their performance over time without being explicitly programmed for the task.
Model: In machine learning and AI, a model is the output of a learning algorithm run on data. A model represents what was learned by a machine learning algorithm. The model is the "thing" that is saved after running a machine learning algorithm on training data and represents the rules, numbers, and any other algorithm-specific data structures required to make predictions.
Multi-Layer Perceptron (MLP): A class of feedforward artificial neural network (ANN) that consists of at least three layers of nodes: an input layer, a hidden layer, and an output layer. MLP utilises a supervised learning technique called backpropagation for training, useful in complex problems like image recognition and natural language processing.
Markov Decision Process (MDP): A mathematical framework used for modelling decision-making situations where outcomes are partly random and partly under the control of a decision-maker. MDPs are crucial for understanding reinforcement learning and for solving various optimisation problems.
N:
Natural Language Processing (NLP): A field of AI that focuses on the interaction between computers and humans through natural language. The ultimate objective of NLP is to read, decipher, understand, and make sense of human languages in a manner that is valuable, enabling tasks like translation, sentiment analysis, and topic extraction.
Neural Network: A series of algorithms that endeavours to recognise underlying relationships in a set of data through a process that mimics the way the human brain operates. Neural networks can adapt to changing input, so the network generates the best possible result without needing to redesign the output criteria.
Naïve Bayes Classifier: A simple probabilistic classifier based on applying Bayes' theorem with strong (naive) independence assumptions between the features. It is highly scalable and effective for large datasets, especially for text classification problems.
Normalisation: A preprocessing technique used in machine learning to adjust the values in the dataset to a common scale without distorting differences in the ranges of values. Normalisation is important for many algorithms to perform well.
Non-Parametric Models: Models that do not make strong assumptions about the form of the mapping function from inputs to outputs. Non-parametric does not mean that such models completely lack parameters but that the number and nature of the parameters are flexible and determined from data.
O:
Overfitting: A modelling error in machine learning that occurs when a function is too closely fit to a limited set of data points. Overfitting the model essentially learns the "noise" in the training data instead of the actual signal, which negatively impacts the model's ability to generalise to new data.
Object Recognition: A computer vision technique for identifying objects in images or videos. Object recognition algorithms leverage machine learning or deep learning to detect the presence of specific objects within an image and classify them into one of many predefined categories or classes.
Optimisation: In the context of machine learning, optimisation refers to the process of adjusting the parameters of algorithms to minimise or maximise some aspect of the data or model, such as error rates or prediction accuracy. Optimisation is crucial for finding the most effective models.
P:
Perceptron: A type of artificial neuron or the simplest form of a neural network, which is capable of binary classification tasks. The perceptron receives multiple input signals, processes them, and produces a single binary output based on a threshold.
Precision: In the context of classification tasks in machine learning, precision is a metric that measures the accuracy of the positive predictions made by the model. It is the ratio of true positive predictions to the total positive predictions (including false positives).
Predictive Analytics: The use of data, statistical algorithms, and machine learning techniques to identify the likelihood of future outcomes based on historical data. It is widely used in marketing, financial services, and operations management.
Principal Component Analysis (PCA): A statistical procedure that utilises orthogonal transformation to convert a set of observations of possibly correlated variables into a set of values of linearly uncorrelated variables called principal components. PCA is used in exploratory data analysis and for making predictive models.
Probabilistic Graphical Models (PGMs): A framework for modelling complex multivariate relationships using probability distributions. PGMs represent the conditional dependencies between random variables through a graph, aiding in inference and decision-making processes.
Python: A high-level, interpreted programming language known for its readability and versatility. Python is one of the most popular languages for AI and machine learning development due to its extensive libraries and frameworks like TensorFlow, PyTorch, and scikit-learn.
Pattern Recognition: The automated recognition of patterns and regularities in data. Pattern recognition is closely related to artificial intelligence and machine learning and is used in a wide range of applications, including speech recognition, image analysis, and biometric identification.
Pooling: A function used in convolutional neural networks to reduce the spatial size of the representation, to decrease the amount of parameters and computation in the network. Pooling layers summarise the features present in regions of the feature map generated by convolutions.
Parameter Tuning: The process of selecting the optimal parameters for a machine learning model to maximise its performance. Parameter tuning involves experimenting with different settings and using validation data to evaluate the model's accuracy.
Quantisation: In digital signal processing and deep learning, quantisation refers to the process of approximating a continuous range of values (or a very large discrete set of values) with a relatively small set of discrete symbols or integer values. Quantisation is essential for compressing models and reducing the computational and memory footprint of AI applications.
Q-Learning: A model-free reinforcement learning algorithm that learns the value of an action in a particular state. It uses a Q-function to measure the quality of an action taken in a given state, and it updates its values based on the reward received after performing an action.
Query: In the context of databases and search engines, a query is a request for information or action. In AI and NLP applications, queries are processed to retrieve information, generate responses, or trigger actions based on the content of the query.
Quasi-Experiment: A research design that resembles an experimental design but lacks the key ingredient of random assignment to treatment or control groups. In AI, quasi-experiments can be used to infer causality from observational data when controlled experiments are not feasible.
Q:
Quantum Computing: A type of computing that takes advantage of the quantum states of subatomic particles to store information. Quantum computers are fundamentally different from binary digital electronic computers and have the potential to perform complex calculations more efficiently. In AI, quantum computing could revolutionise the speed and complexity of data processing for machine learning algorithms.
Q-Lerning: A form of model-free reinforcement learning where an agent learns to achieve a goal by taking actions in an environment and receiving rewards or penalties. The agent learns a policy, mapping states of the environment to actions that maximise cumulative reward.
Query Optimisation: In databases and information retrieval, query optimisation involves modifying a query to improve the efficiency of retrieving data. In AI, query optimisation can be applied to speed up data retrieval processes for machine learning tasks, especially when dealing with large datasets.
Quantisation (in AI): The process of reducing the precision of the weights and, optionally, activations of models to reduce memory and improve computational efficiency. In deep learning, quantisation techniques are essential for deploying models on resource-constrained devices.
Quadratic Loss Function: A type of loss function used in optimisation and machine learning algorithms. It measures the square of the difference between the predicted values and the actual values, emphasising larger errors more heavily than smaller errors.
Qualitative Analysis: In AI, qualitative analysis refers to the examination of non-quantifiable elements of data, such as text or images, to understand patterns, themes, or meanings. Techniques like natural language processing and image recognition can automate qualitative analysis tasks.
Quantum Machine Learning: An emerging field at the intersection of quantum computing and machine learning. It explores how quantum algorithms can be used to improve machine learning tasks, potentially offering exponential speedups for certain computations.
Quasi-Newton Methods: Optimisation algorithms used in training machine learning models, especially in the context of deep learning. These methods approximate the Hessian matrix, which represents the second-order partial derivative of a function, to find the minimum or maximum of a loss function more efficiently than standard gradient descent.
Query Language: In AI systems that interact with databases or structured data sources, a query language is used to retrieve information based on specific criteria. Examples include SQL for relational databases and SPARQL for querying RDF data in semantic web applications.
Qubit: The basic unit of quantum information in quantum computing, analogous to the binary digit (bit) in classical computing. Qubits have properties that allow them to represent and process more information than bits, which is why they are considered for enhancing AI algorithms.
R:
Reinforcement Learning (RL): A type of machine learning technique that enables an agent to learn in an interactive environment by trial and error using feedback from its own actions and experiences. RL is widely used in game playing, robotics, and navigation systems.
Random Forest: An ensemble learning method for classification, regression, and other tasks that operates by constructing a multitude of decision trees at training time. For classification tasks, the output of the random forest is the class selected by most trees.
Recall: In the context of information retrieval and machine learning, recall is a metric that measures the ability of a model to identify all relevant instances within a dataset. It is the ratio of the true positive rate to the number of all relevant samples.
Recurrent Neural Network (RNN): A class of artificial neural networks where connections between nodes form a directed graph along a temporal sequence. This allows it to exhibit temporal dynamic behaviour. Unlike feedforward neural networks, RNNs can use their internal state (memory) to process sequences of inputs.
Regression Analysis: A statistical method for estimating the relationships among variables. It is widely used for prediction and forecasting, where its use has substantial overlap with the field of machine learning.
Regularisation: A technique used to prevent overfitting by adding a penalty on the magnitude of model parameters or coefficients. Regularisation techniques such as L1 and L2 regularisation are essential in the training of models that generalise well to unseen data.
Robotics: An interdisciplinary branch of engineering and science that includes mechanical engineering, electronic engineering, information engineering, computer science, and others. Robotics deals with the design, construction, operation, and use of robots, as well as computer systems for their control, sensory feedback, and information processing.
Rule-based System: A type of software system that uses rules as the knowledge representation. It makes decisions based on predefined logical rules rather than patterns or historical data. Rule-based systems are often used in expert systems and other applications requiring complex decision-making.
Representation Learning: A set of techniques in machine learning where a system can automatically discover the representations needed for feature detection or classification from raw data. This learning methodology allows a machine to be fed with raw data and to discover the representations necessary for detection or classification.
Robotic Process Automation (RPA): The technology that allows anyone today to configure computer software, or a “robot” to emulate and integrate the actions of a human interacting within digital systems to execute a business process. RPA robots utilise the user interface to capture data and manipulate applications just like humans do.
ReLU (Rectified Linear Unit): A type of activation function that is defined as the positive part of its argument. Where an input is positive, ReLU returns the value, but for any negative input, ReLU returns zero. It has become the default activation function for many types of neural networks because a model that uses it is easier to train and often achieves better performance.
S:
Supervised Learning: A type of machine learning algorithm that is trained on labelled data, meaning the algorithm is provided with input-output pairs. The goal is to learn a mapping from inputs to outputs, making it possible to predict the output for new, unseen inputs.
Support Vector Machine (SVM): A supervised machine learning model used for classification and regression tasks. It works by finding the hyperplane that best separates different classes in the feature space.
Semantic Analysis: In natural language processing (NLP), semantic analysis is the process of understanding the meaning and interpretation of words, phrases, and sentences in the context of the language. It involves discerning the structures of linguistic significance to understand the nuances of meaning.
Stochastic Gradient Descent (SGD): An optimisation technique used in machine learning and deep learning for minimising the loss function. It is a variant of gradient descent where updates to the parameters are made using a subset of the training data, chosen randomly, making the method more computationally efficient.
Sequence Modelling: A type of model in machine learning designed to predict or generate sequences of data, such as time series data, sentences, or musical notes. Recurrent neural networks (RNNs) and Long Short-Term Memory (LSTM) networks are commonly used for sequence modelling.
Softmax Function: A function that takes as input a vector of K real numbers and normalises it into a probability distribution consisting of K probabilities. It is often used in the final layer of a neural network-based classifier to ensure the output values are in the range (0, 1) and sum up to 1.
Sentiment Analysis: Also known as opinion mining, it is a natural language processing technique used to determine whether data is positive, negative, or neutral. Sentiment analysis is widely used in social media monitoring, market research, and customer service.
Sparse Data: Data sets that contain a large number of elements that are zero or otherwise do not contain useful information. Sparse data presents challenges for certain types of machine learning algorithms, which may perform better with dense, or non-sparse, data.
Speech Recognition: The ability of a machine or program to identify words and phrases in spoken language and convert them into a machine-readable format. It is a critical component of natural language processing systems.
Swarm Intelligence: A field of artificial intelligence that is inspired by the collective behaviour of decentralised, self-organised systems, such as ant colonies or bird flocking. Swarm intelligence principles are used to develop algorithms for optimisation, robotics, and problem-solving.
Synthetic Data: Artificially generated data that is not obtained by direct measurement, often used for training machine learning models where real data may be scarce or sensitive. Synthetic data must be representative enough to ensure that models trained on it perform well on real data.
T:
Transfer Learning: A research problem in machine learning that focuses on storing knowledge gained while solving one problem and applying it to a different but related problem. This is particularly useful in deep learning where pre-trained models are adapted for new tasks.
TensorFlow: An open-source machine learning framework developed by Google Brain. It provides a comprehensive ecosystem of tools, libraries, and community resources that allows researchers to push the state-of-the-art in ML, and developers to easily build and deploy ML-powered applications.
Tokenisation: In natural language processing (NLP), tokenisation is the process of breaking down text into smaller units called tokens, which can be words, characters, or subwords. It is a fundamental step in preprocessing text data for machine learning models.
Time Series Analysis: A statistical technique that deals with time series data, or trend analysis, which consists of sequences of data points recorded or measured at successive time intervals. Time series analysis is used for forecasting and predicting future values based on previously observed values.
Tree-Based Methods: Machine learning methods that involve the use of decision trees or ensembles of decision trees (like Random Forests and Gradient Boosting Machines) to make predictions or classify data. These methods are popular due to their interpretability and effectiveness across various tasks.
Text Mining: The process of deriving high-quality information from text through the identification of patterns and trends via means such as statistical pattern learning. Text mining is widely used in business intelligence, research, and web mining.
True Positive/Negative: In the context of binary classification in machine learning, a true positive is an outcome where the model correctly predicts the positive class. Similarly, a true negative is an outcome where the model correctly predicts the negative class.
Transformer Models: A type of deep learning model introduced in the paper "Attention is All You Need," which uses self-attention mechanisms and has significantly improved the performance of many natural language processing tasks. Examples include BERT, GPT, and other models based on the Transformer architecture.
Thompson Sampling: A method used in multi-armed bandit problems and reinforcement learning for balancing the exploration-exploitation trade-off. It involves choosing actions based on probability matching, which leads to an optimal balance over time.
Turing Test: A test of a machine's ability to exhibit intelligent behaviour equivalent to, or indistinguishable from, that of a human. Proposed by Alan Turing in 1950, the Turing Test is considered one of the seminal ideas in the philosophy of artificial intelligence.
Topic Modelling: A type of statistical model for discovering abstract topics that occur in a collection of documents. Topic modelling is a frequently used text-mining tool for discovery of hidden semantic structures in a text body, used in natural language processing and machine learning.
U:
Unsupervised Learning: A type of machine learning algorithm used to draw inferences from datasets consisting of input data without labelled responses. The most common unsupervised learning method is cluster analysis, which is used for exploratory data analysis to find hidden patterns or grouping in data.
Universal Approximation Theorem: A theorem stating that a feed-forward network with a single hidden layer containing a finite number of neurons can approximate continuous functions on compact subsets of R^n, under mild assumptions on the activation function. This theorem is foundational in demonstrating the capability of neural networks to represent a wide variety of functions.
Underfitting: A modelling error in machine learning where a model is too simple to capture the underlying structure of the data. An under-fitted model performs poorly on both the training data and unseen data, indicating it has not learned the relevant patterns in the training data.
Utility Theory: In AI and economics, utility theory is a decision-making process that models an agent's preferences concerning a set of outcomes or goods. Utility theory helps in understanding how decisions are made based on the perceived utility or satisfaction from a particular choice.
User Interface (UI) for AI: The means by which a human interacts with a computer, machine, or application. AI-enhanced user interfaces use machine learning and natural language processing to create more intuitive and user-friendly experiences, often allowing users to interact with systems in more natural and human-like ways.
Unstructured Data: Information that either does not have a predefined data model or is not organised in a pre-defined manner. Unstructured data is typically text-heavy, but may contain data such as dates, numbers, and facts as well. Handling unstructured data is a common task in natural language processing and other AI fields.
V:
Validation Set: A subset of data used to assess the performance of a model during the training phase, but not used for training the model. The validation set helps in tuning the model's parameters and provides an unbiased evaluation of a model fit on the training dataset.
Vectorisation: The process of converting non-numeric data into a numeric format so that it can be used in machine learning algorithms. In natural language processing, vectorisation often involves converting text into vectors of numbers based on various schemes like Bag of Words, TF-IDF, or word embeddings.
Vision Systems: In AI, vision systems are designed to interpret and understand the visual world. These systems use techniques from machine learning and computer vision to recognise objects, faces, scenes, and activities in images and videos
Variational Autoencoder (VAE): A type of autoencoder neural network introduced to learn deep representations of data in an unsupervised manner. VAEs are particularly useful for generative tasks, where they can generate new data points that are similar to the input data.
Virtual Agents: AI systems designed to interact with humans through natural language. Virtual agents, also known as chatbots or digital assistants, can understand and respond to user queries, perform tasks, and provide assistance or information.
Viterbi Algorithm: A dynamic programming algorithm for finding the most likely sequence of hidden states—called the Viterbi path—that results in a sequence of observed events, especially in the context of Markov information sources and hidden Markov models (HMM).
Voice Recognition: The ability of a machine or program to receive and interpret dictation or understand and carry out spoken commands. Voice recognition technology is a key component of AI systems that interact with users via spoken language.
Volatile Memory: Computer memory that requires power to maintain the stored information. In the context of AI hardware, volatile memory (such as RAM) is used for processing and temporarily storing data while models are being trained or inference is being made.
Voronoi Diagram: A partitioning of a plane into regions based on distance to points in a specific subset of the plane. In AI and machine learning, Voronoi diagrams can be used for clustering analysis and to visualise the structure of datasets.
VGG Network: A deep convolutional neural network architecture named after the Visual Geometry Group at the University of Oxford. VGG networks have been influential in the field of deep learning, particularly in image recognition tasks.
W:
Weight: In the context of neural networks and machine learning, weights are parameters that determine the strength of the connection between two neurons. Adjusting these weights during the training process is how the network learns to make accurate predictions or classifications.
Word Embedding: A type of word representation that allows words to be represented as vectors in a continuous vector space. This technique enables capturing the context of a word in a document, its semantic and syntactic similarity, relation with other words, etc. Word embeddings are used extensively in natural language processing (NLP) tasks.
Weak AI: Also known as narrow AI, refers to artificial intelligence systems that are designed and trained for a specific task. Weak AI is in contrast to strong AI, which refers to AI systems with generalised human cognitive abilities.
Workflow Automation: The design, execution, and automation of processes based on workflow rules where human tasks, data or files are routed between people or systems based on pre-defined business rules. In AI, workflow automation is often enhanced with machine learning algorithms to improve efficiency and decision-making.
Watson: IBM Watson is a computer system capable of answering questions posed in natural language, developed in IBM's DeepQA project. Watson combines advanced natural language processing, information retrieval, knowledge representation, automated reasoning, and machine learning technologies.
Wide Learning: A machine learning concept that involves creating a broad set of linear models or rules that are combined into a single model. Wide learning is particularly useful for memorisation and managing data with a large number of sparse features, often used in conjunction with deep learning models for better generalisation.
Wrapper Method: In feature selection, wrapper methods evaluate subsets of variables to determine which features result in the highest performing model. Unlike filter methods that select features based on their relationships with the target variable, wrapper methods use the model's performance as the evaluation criterion.
Web Scraping: The process of extracting data from websites. In AI and machine learning projects, web scraping can be used to gather large datasets from the Internet, which are then used for training models, especially in fields like NLP and market analysis.
White Box Model: A machine learning model whose inner workings are well understood and can be easily explained. This is in contrast to black-box models, where the decision-making process is not transparent. White box models are important for applications requiring interpretability and accountability.
Weka: A collection of machine learning algorithms for data mining tasks, written in Java and developed at the University of Waikato, New Zealand. Weka contains tools for data pre-processing, classification, regression, clustering, association rules, and visualisation, making it a popular choice for academic and educational purposes.
Weight Decay: A regularisation technique that adds a penalty on the size of network weights to the loss function. By discouraging large weights, weight decay helps prevent overfitting and makes the model simpler and less prone to overfitting on the training data.
Weak AI
Weak AI, also known as Narrow AI, is AI that is designed and trained for a particular task. Virtual personal assistants, such as Apple's Siri, are a form of weak AI.
X:
XAI (Explainable Artificial Intelligence): A set of processes and methods that allows human users to understand and trust the results and output created by machine learning algorithms. XAI aims to make AI decisions more transparent and explainable without sacrificing performance.
XML (eXtensible Markup Language): A markup language that defines a set of rules for encoding documents in a format that is both human-readable and machine-readable. In AI and machine learning, XML is often used for data representation, configuration files, and communication protocols between different systems.
XGBoost (eXtreme Gradient Boosting): An open-source software library that provides an efficient and scalable implementation of gradient boosted decision trees. XGBoost is widely used in machine learning for classification, regression, and ranking problems due to its speed and performance.
XOR Problem: In the context of neural networks, the XOR problem is a classic problem that cannot be solved by a single layer perceptron. It involves creating a model that correctly classifies the outputs of an XOR gate, demonstrating the need for multi-layer networks to solve nonlinear problems.
X-axis in Machine Learning Models: In the graphical representation of data or models, the x-axis often represents the input variables or features, while the y-axis represents the output or predictions. Understanding the relationship between the two can help in interpreting the model's behaviour.
XPath: A query language that allows for the navigation of XML documents in order to select nodes or compute values. In AI applications involving data extraction and processing, XPath can be used to parse and extract information from XML-based data sources.
Xception: A deep convolutional neural network architecture inspired by Inception, where the Inception modules have been replaced with depth wise separable convolutions. Xception stands for "Extreme Inception," and it's designed to improve efficiency and performance in image classification tasks.
XAI Techniques: Methods and approaches used within explainable AI to make machine learning models more interpretable, such as feature importance scores, decision trees, and visual explanations. These techniques aim to shed light on how models make decisions, helping to build trust and understanding.
X-means Clustering: An extension of the k-means algorithm that automatically determines the optimal number of clusters. X-means clustering starts with a predefined number of clusters and iteratively splits them to improve the fit, using a criterion like the Bayesian Information Criterion (BIC) to decide when to stop.
Xenobots: The term "xenobots" refers to programmable biological robots, typically derived from frog cells. While not directly related to conventional AI, the creation and programming of xenobots involve principles of design and control that intersect with artificial intelligence and robotics, showcasing a novel approach to understanding life and machinery.
Y:
Yield Prediction: In agriculture and farming, yield prediction involves using machine learning models to forecast the amount of crop that will be produced in a given season. These predictions can be based on various factors, including weather data, soil quality, crop type, and farming practices, helping farmers make informed decisions.
YOLO (You Only Look Once): A popular deep learning algorithm for real-time object detection that frames object detection as a regression problem to spatially separated bounding boxes and associated class probabilities. YOLO is known for its speed and accuracy in detecting objects in images or video streams.
Y-axis in Machine Learning Models: In the graphical representation of data or model outputs, the y-axis typically represents the predicted values or outputs of the model, while the x-axis represents the input features or variables. The relationship between the two axes helps in visualising and interpreting the model's predictions.
YARN (Yet Another Resource Negotiator): A cluster management technology in the Hadoop ecosystem that allows for the resource management and job scheduling of distributed applications. YARN is significant in AI and big data processing for managing computing resources in large-scale environments.
Yellow Box Testing: A testing methodology that combines both black-box and white-box testing approaches. In the context of AI and machine learning, yellow box testing might involve examining the model's performance (black-box) and its internal workings or logic (white-box) to ensure accuracy and reliability.
Yottabyte: A unit of data measurement that is equal to 2^80 bytes. In the realm of big data and AI, dealing with yottabytes of data represents significant challenges in terms of data storage, processing, and analysis, pushing the boundaries of current technologies and algorithms.
Yield Optimisation: In digital advertising and online publishing, yield optimisation uses algorithms to maximise the revenue generated from ad space. AI techniques analyse data on user behaviour, advertiser bids, and content performance to dynamically adjust which ads are shown to maximise engagement and revenue.
Y Combinator: A well-known startup accelerator that has funded several AI and machine learning startups among its cohorts. Y Combinator provides seed funding, guidance, and resources to startups, helping them refine their business models and technologies, including those focused on AI innovations.
YUV: A colour space used in video compression and processing, where Y represents the luminance component, and U and V are the chrominance components. In AI-based video analysis and processing, understanding and manipulating the YUV colour space can be crucial for tasks such as object detection, facial recognition, and video enhancement.
Z:
Zero-shot Learning: A machine learning technique where a model learns to correctly make predictions for tasks it has not explicitly seen during training. Zero-shot learning aims to improve the model's generalisation ability to new, unseen categories.
Z-score Normalisation: Also known as standard score or z-score scaling, it is a technique used in data preprocessing where the values of a feature are normalised based on the feature's mean and standard deviation. Z-score normalisation ensures that each feature contributes equally to the distance computations in algorithms.
Zigbee: A specification for a suite of high-level communication protocols used to create personal area networks with small, low-power digital radios. Though not directly an AI technology, Zigbee-enabled devices often incorporate AI and machine learning for smart home and IoT applications.
Z-buffering: A computer graphics technique for rendering depth in 3D models by storing and updating the depth information of each pixel on the screen. While not specific to AI, z-buffering techniques are used in virtual reality (VR) and augmented reality (AR) applications that leverage AI for more immersive experiences.
ZooKeeper: An open-source server which enables highly reliable distributed coordination. In the context of distributed machine learning and AI, ZooKeeper can manage and coordinate the distributed processes across large-scale clusters.
Zeroth Law of Robotics: A precept introduced by the science fiction writer Isaac Asimov, stating that a robot may not harm humanity, or, by inaction, allow humanity to come to harm, taking precedence over his earlier three laws of robotics. This concept, while fictional, stimulates discussion on the ethics and governance of AI and robotics.
Zeta Architecture: A conceptual architecture that extends beyond the lambda and kappa architectures used for processing big data. It is designed to handle an organisation's entire data processing pipeline, from real-time data processing to batch processing analytics, which can include AI and machine learning workflows.
Zone of Proximal Development (ZPD): A concept from educational psychology and learning theory which refers to the difference between what a learner can do without help and what they can achieve with guidance and encouragement from a skilled partner. In AI, this concept can apply to systems designed to adapt and personalise learning experiences.
ZSL (Zero-Shot Learning): See "Zero-shot Learning" for a detailed definition. This approach is particularly promising for AI applications where the cost of labelling data is prohibitive or when dealing with highly dynamic environments where new classes of data frequently emerge.
This glossary provides a foundational understanding of essential AI terms, serving as a springboard for deeper exploration into this transformative field. As AI continues to evolve, so too will this lexicon, reflecting the relentless pace of innovation in artificial intelligence.
