A

​

​

AI Agent 

A software entity or system capable of perceiving its environment, making decisions, and taking actions to achieve specific goals. AI agents operate autonomously or semi-autonomously and can be used in applications like chatbots, virtual assistants, and robotics.
 

Algorithm

A structured set of rules or procedures that a computer follows to perform a task or solve a problem. Algorithms are foundational to AI, enabling machines to process data, recognize patterns, and make predictions efficiently.
 

Artificial General Intelligence (AGI)

A highly advanced form of AI capable of performing any intellectual task that a human can do. AGI can generalize knowledge across different domains, demonstrating reasoning, problem-solving, and adaptability similar to human intelligence.
 

Artificial Intelligence (AI)

The branch of computer science focused on building systems that can mimic human cognitive functions such as learning, reasoning, perception, and decision-making. AI is applied in diverse areas, from autonomous vehicles to language processing.
 

Artificial Neural Network (ANN)

A machine learning model inspired by the structure of the human brain, consisting of layers of interconnected nodes (neurons). ANNs process information by passing data through these nodes, making them effective for tasks like image recognition and natural language processing.
 

Attention Mechanism

A technique in neural networks that allows the model to focus on specific parts of the input data, assigning higher importance to relevant elements. This mechanism improves performance in tasks like translation, summarization, and question-answering.
 

Augmented Intelligence

A human-centred approach to AI that aims to enhance, rather than replace, human intelligence. By combining machine capabilities with human expertise, augmented intelligence supports better decision-making and problem-solving in complex scenarios.
 

Autonomous Systems

Systems are capable of operating independently by perceiving their environment, making decisions, and executing tasks without human intervention. Autonomous systems are used in areas like self-driving cars, drones, and industrial automation.

 

B​

​

​

Bias (in AI)

A systematic error in AI models that causes unfair or skewed outcomes. Bias can arise from unbalanced datasets, flawed algorithms, or human biases in data collection, leading to ethical and performance issues.
 

Big Data

Extremely large and complex datasets that are difficult to process using traditional methods. Big data is characterized by the "3 Vs" (Volume, Velocity, and Variety) and is foundational to many AI applications that rely on massive amounts of data for training.
 

Biometric Authentication

The use of biological characteristics (such as fingerprints, facial recognition, or voice patterns) to verify a person's identity. AI-powered biometric systems enhance security and usability in applications like smartphones and access control.
 

Black Box Model

A term used to describe AI models, particularly deep learning models, whose internal workings are not easily interpretable by humans. These models can make accurate predictions but often lack transparency in explaining how decisions are made.
 

Bot

A software application designed to automate repetitive tasks or simulate human interaction. AI-powered bots, such as chatbots or virtual assistants, can process natural language to perform tasks like answering questions or providing customer support.
 

Brain-Computer Interface (BCI)

A system that enables direct communication between a user’s brain and an external device, often using AI to interpret neural signals. BCIs have applications in medical rehabilitation and assistive technologies.

​
 

C

​​​

​

Chain-of-Thought (CoT)

A reasoning approach in large language models that involves breaking down problems into intermediate steps to improve the quality and interpretability of the model's responses. It is particularly effective in complex tasks like math or logic.

 

Chatbot

An AI-powered software application that simulates human conversation using natural language processing. Chatbots are used for customer service, virtual assistance, and automating repetitive tasks.

 

Cloud Computing

The delivery of computing resources (such as storage, processing power, and AI tools) over the internet. Cloud platforms like AWS, Azure, and Google Cloud provide scalable infrastructure for deploying AI models and applications.

 

Clustering

A type of unsupervised learning where data points are grouped into clusters based on their similarity. Common algorithms include K-Means, DBSCAN, and Hierarchical Clustering, often used in market segmentation or image compression.

 

Cold Start Problem

A challenge in recommendation systems where insufficient data about a new user or item makes it difficult to provide accurate recommendations. Solutions often include hybrid approaches or content-based filtering.

 

Collaborative Filtering

A recommendation technique that suggests items to users based on the preferences of other users with similar interests. It is widely used in recommendation systems like Netflix and Amazon.

 

Computer Vision

A field of AI that enables machines to interpret and analyze visual data from the world, such as images and videos. Applications include facial recognition, object detection, and autonomous vehicles.

 

Convolutional Neural Network (CNN)

A deep learning model specifically designed for processing grid-like data such as images. CNNs use convolutional layers to automatically detect and learn spatial features like edges, shapes, and textures.

​

 

D

​​

​

Data Labeling

The process of annotating raw data with meaningful labels or tags enables supervised learning algorithms to understand and learn from the data. Examples include tagging images with objects or labelling text for the sentiment.

​

Data Mining

The practice of discovering patterns, correlations, and insights in large datasets using statistical and machine learning techniques. It is widely used in fields such as marketing, healthcare, and fraud detection.

​

Data Normalization

A preprocessing technique is used to scale numerical data to a standard range (e.g., 0 to 1) or distribution. It ensures that all features contribute equally to the learning process, improving model performance.

​

Data Science

An interdisciplinary field combining statistics, machine learning, domain expertise, and programming to analyze and extract insights from data. Data science drives decision-making in business, research, and technology.

​

Dataset

A structured collection of data used for training, validating, and testing machine learning models. Datasets can consist of images, text, audio, or numerical values and are crucial for model development.

​

Decision Tree

A machine learning algorithm that splits data into branches based on feature values to make decisions or predictions. Decision trees are interpretable and commonly used for classification and regression tasks.

​

Deep Learning (DL)

A subset of machine learning that uses neural networks with many layers to learn complex patterns and representations in large datasets. It is widely applied in tasks like image recognition, natural language processing, and autonomous systems.

​

​​E

​​

​

Embedding

A numerical representation of data, such as words, sentences, or images, in a lower-dimensional vector space. Embeddings capture relationships and similarities between data points, commonly used in natural language processing and recommendation systems.

 

Encoder-Decoder Architecture

A neural network framework consisting of an encoder that compresses input data into a latent representation and a decoder that reconstructs the original data or generates new data. This architecture is widely used in machine translation and image captioning.

 

Extrapolation

The process of making predictions beyond the range of observed data. While interpolation predicts within known data bounds, extrapolation assumes that patterns in the data extend beyond those bounds, which can be riskier.

​​

​

F

​​

​

Few-Shot Learning

A training approach where models learn to perform tasks with a minimal number of labelled examples. Few-shot learning is particularly useful in scenarios with limited data, such as medical imaging and rare language tasks.

​

Fine-tuning

The process of adjusting a pre-trained model to a specific task by continuing its training on a smaller, task-specific dataset. Fine-tuning leverages pre-learned knowledge and reduces the need for extensive labelled data.

​

Forward Propagation

The process of passing input data through a neural network, layer by layer, to compute predictions. Forward propagation is the initial step in training, followed by backpropagation to adjust the model weights.

​

Framework

A software library or platform designed to simplify the development of AI and machine learning models. Examples include TensorFlow, PyTorch, and Scikit-learn, which provide tools for model building, training, and deployment.

​

Foundation AI Models

Large, pre-trained models designed to serve as a base for fine-tuning across a wide range of tasks. They are capable of general-purpose understanding and generation, forming the backbone of modern AI applications.

​​

​

G

​

​

GAN (Generative Adversarial Network)

A deep learning framework consisting of two neural networks - a generator and a discriminator - competing against each other. The generator creates synthetic data, while the discriminator evaluates its authenticity, resulting in high-quality data generation.

​

Graph Neural Network (GNN)

A type of neural network designed to work with graph-structured data, such as social networks or molecular structures. GNNs capture relationships between nodes and edges for tasks like node classification and link prediction.

​

Generative Model

A type of model that learns the underlying distribution of a dataset to generate new data samples. Examples include GANs, Variational Autoencoders (VAEs), and language models like GPT.

​

Generative Artificial Intelligence (GenAI)

AI systems are designed to create new content, such as text, images, music, or code, often based on patterns learned from existing data. Applications include ChatGPT, DALL·E, and automated design tools.

​

Generalization

The ability of a machine learning model to perform well on unseen data by learning patterns that are not specific to the training set. Overfitting hampers generalization, while regularization and cross-validation enhance it.

​​

​

H

​​

​

Hallucination

A phenomenon in generative AI models where the output includes incorrect or fabricated information that is not present in the input or training data. It is a key challenge in ensuring AI reliability.

​

Hyperparameter

A parameter in a machine learning model that is not learned during training but is set before the learning process. Examples include learning rate, number of layers, and batch size. Hyperparameter tuning optimizes these values for better performance.

 

Human-in-the-Loop (HITL)

A design approach where humans actively participate in the AI training, validation, or decision-making process. HITL ensures better performance, interpretability, and fairness in AI systems.

 

Human-Centered AI

A paradigm that focuses on designing AI systems to augment human capabilities and prioritize user needs. It emphasizes transparency, usability, and ethical considerations.

 

Hybrid Cloud

A computing environment that combines on-premises infrastructure with cloud-based resources. In AI, hybrid clouds are often used for scalable training and deployment while maintaining control over sensitive data.

​​

​

I

​

​

Image Recognition

A computer vision task that involves identifying and classifying objects, scenes, or patterns within an image. Applications include facial recognition, medical imaging, and autonomous vehicles.

 

Inference

The process of using a trained machine learning model to make predictions or decisions based on new, unseen data. Inference is the application phase of a model after it has been trained.

 

Interpretability

The degree to which a human can understand the reasoning and decision-making process of an AI model. Techniques like feature importance, LIME, and SHAP enhance interpretability in complex models.

 

Intelligent Agent

An autonomous entity that perceives its environment, processes data and takes actions to achieve specific goals. Intelligent agents are used in applications like virtual assistants, robotics, and autonomous vehicles.

 

Internet of Things (IoT)

A network of interconnected devices that collect and exchange data, often with the help of AI for processing and decision-making. IoT applications include smart homes, industrial automation, and healthcare monitoring.

​​

​

J

 

​

Joint Training

A training strategy where multiple tasks or models are trained simultaneously, sharing representations to improve performance. It is often used in multitask learning and transfer learning.

 

JSON (JavaScript Object Notation)

A lightweight data format widely used for data exchange in AI applications, especially in APIs and model deployment. JSON is easy to parse and read by both humans and machines.

​​

​

K

​

 

k-Means Clustering

An unsupervised learning algorithm that partitions a dataset into kkk clusters based on the similarity of data points. It iteratively adjusts cluster centroids and assignments to minimize within-cluster variance.

 

Keypoint Detection

A computer vision technique that identifies distinctive points in an image, such as corners or edges. It is often used in object recognition, image

 

Knowledge Representation

The field of AI focuses on designing ways to represent and organize information so that machines can reason about it. Techniques include semantic networks, ontologies, and logical representations.

​​

​

L

 

​

Label

A value or category is assigned to data points in supervised learning to indicate the desired output. Labels are essential for training models to recognize patterns and make accurate predictions.

 

Latency

The time delay between an input being provided to a system and the system’s response. In AI, latency is critical in real-time applications like autonomous vehicles, where low latency ensures timely decisions.

 

Layer

A building block of neural networks, where data is processed and transformed through weighted connections and activation functions. Layers can be categorized as input, hidden, or output layers.

 

Learning Rate

A hyperparameter that controls the step size of weight updates during model training. Choosing an appropriate learning rate is crucial for ensuring convergence and avoiding overshooting or slow progress.

 

Loss Function

A mathematical function that quantifies the difference between a model’s predictions and the actual labels. The goal of training is to minimize the loss function to improve model accuracy.

 

Language Model

A model that predicts the likelihood of sequences of words or characters. Language models, such as GPT and BERT, are foundational to tasks like text generation, translation, and sentiment analysis.

 

Large Language Model (LLM)

A type of AI model trained on vast amounts of text data to perform natural language processing tasks like text generation, summarization, and translation. Examples include GPT, BERT, and LLaMA.

​​

​

M

 

​

Machine Learning (ML)

A subset of AI that focuses on creating algorithms capable of learning patterns from data and making predictions or decisions without being explicitly programmed.

 

Model Interpretability

The ability to understand and explain how a machine learning model makes decisions. Techniques like LIME, SHAP, and feature importance enhance interpretability in complex models.

 

Multimodal Learning

An approach that combines and processes data from multiple modalities, such as text, images, and audio. Applications include video captioning and speech recognition.

 

Multimodal AI Model

A model that integrates and processes data from multiple modalities, such as text, images, audio, or video, to perform tasks. Examples include OpenAI's GPT-4 with multimodal capabilities.

​​

​

N

 

​

Named Entity Recognition (NER)

A natural language processing task that identifies and classifies entities in text, such as names, dates, locations, and organizations. NER is a foundational component of information extraction systems.

 

Natural Language Processing (NLP)

A branch of AI that enables machines to understand, interpret, and generate human language. Common applications include machine translation, sentiment analysis, and text summarization.

 

Neural Network

A machine learning model inspired by the structure of the human brain, consisting of layers of interconnected nodes (neurons). Neural networks are the foundation of deep learning.

 

Node

A basic unit in a neural network that processes input data and applies a transformation, such as a weighted sum followed by an activation function. Nodes are also used to represent entities in graph data structures.

 

Normalization

A preprocessing technique that adjusts data to a standard scale, improving numerical stability and ensuring features contribute equally to model training. Common methods include Min-Max scaling and Z-score normalization.

 

No-Code AI Platform

A development environment that allows users to build, deploy, and manage AI models without requiring programming skills. These platforms provide drag-and-drop interfaces and pre-built components to democratize AI adoption.

​​

​

O

 

​

One-Shot

A machine learning approach where a model learns to perform a task with only one labelled example. This is especially useful in scenarios with limited data, such as rare language or visual concepts.

 

Online Learning

A machine learning approach where the model is updated incrementally as new data becomes available, rather than being trained on a static dataset. It is useful in dynamic environments like stock trading or real-time personalization.

 

Overfitting

A situation where a machine learning model performs well on training data but poorly on unseen data due to excessive focus on noise or specific patterns in the training set. Regularization and cross-validation help prevent overfitting.

 

Outlier

A data point that significantly deviates from the overall pattern of the dataset. Outliers can distort model training and are often addressed through techniques like anomaly detection or robust statistics.

 

Open-Source

A development model where the source code of software, frameworks, or models is made publicly available for anyone to view, modify, and distribute. In AI, open-source projects like TensorFlow, PyTorch, and Hugging Face have accelerated innovation by enabling collaborative research and development.

​

​

P

 

​

Parameter

A variable in a machine learning model is learned from data during training, such as weights in neural networks. Parameters define how the model transforms input data into predictions.

 

Principal Component Analysis (PCA)

A dimensionality reduction technique that transforms data into a smaller set of uncorrelated variables called principal components. PCA is used for visualization and noise reduction.

 

Prompting

A method used in large language models (LLMs) where specific instructions or context are provided as input to guide the model’s output. Prompting enables the customization of responses without requiring changes to the underlying model.

 

Propagation

The process of passing data through a machine learning model. Forward propagation is used to make predictions, while backpropagation adjusts parameters to minimize errors.

 

Python

A high-level programming language widely used in AI and machine learning for its simplicity and rich ecosystem of libraries like TensorFlow, PyTorch, and Scikit-learn.

 

Predictive Model

A machine learning model designed to predict future outcomes based on historical data. Predictive models are widely applied in fields like finance, healthcare, and marketing.

 

Pretrained Model

A machine learning model that has been trained on a large dataset and can be fine-tuned for specific tasks. Examples include BERT for natural language processing and ResNet for image recognition.

​​

​

Q

 

​

Quantum Computing

A computing paradigm that leverages quantum mechanics to perform calculations. Quantum computing holds the potential for solving problems in AI that are computationally infeasible for classical computers, such as large-scale optimization and cryptography.

​​

​

R

 

​

Random Forest

An ensemble learning algorithm that combines multiple decision trees, trained on different subsets of data, to improve prediction accuracy and reduce overfitting. It is used for both classification and regression tasks.

 

Recurrent Neural Network (RNN)

A type of neural network designed for sequential data, where connections between nodes form a directed cycle, enabling the model to retain information about previous inputs. RNNs are used in tasks like language modelling and time series forecasting.

 

Regular Expressions (RegEx)

A sequence of characters defines a search pattern used for text matching and manipulation. RegEx is widely used in tasks like data cleaning, information extraction, and natural language processing.

 

Reinforcement Learning (RL)

A machine learning paradigm is where an agent learns to make decisions by interacting with an environment and receiving rewards or penalties. It is widely used in robotics, game-playing, and optimization problems.

 

Regression

A type of supervised learning used to predict continuous outcomes based on input features. Common algorithms include linear regression, polynomial regression, and support vector regression.

 

Reward Function

A function in reinforcement learning that quantifies the immediate feedback the agent receives for taking an action in a given state. The goal of the agent is to maximize cumulative rewards over time.

​​

​

S

 

​

Scalability

The ability of a machine learning model or system to handle increasing amounts of data or computational demand without significant performance degradation.

 

Semantic Search

An advanced search technique that understands the intent and contextual meaning of queries rather than relying solely on keyword matching. It is used in AI-powered search engines and question-answering systems.

 

Stop Words

Common words like "the," "is," and "and" are often removed during natural language processing to focus on meaningful content. Stop word removal is a standard step in text preprocessing.

 

Structured Data

Data that is organized into a well-defined schema, such as tables with rows and columns. It contrasts with unstructured data, like text or images, and is easier to process in machine learning.

 

Supervised Learning

A machine learning paradigm where models are trained on labelled data, learning to map inputs to known outputs. Applications include classification, regression, and predictive modelling.

 

Synthetic Data

Artificially generated data used to augment training datasets or simulate real-world scenarios. Synthetic data is commonly used in privacy-sensitive applications and scenarios with limited real data.

​​

​

T

 

​

Test Data

A subset of data that is separate from the training data, used to evaluate the performance of a trained machine-learning model. It ensures the model generalizes well to unseen data.

 

Token

A basic unit of text, such as a word, subword, or character, used in natural language processing. Tokenization divides text into tokens to prepare it for analysis or modelling.

 

Tokenization

The process of splitting text into smaller units (tokens), such as words, subwords, or characters. It is a crucial preprocessing step in tasks like sentiment analysis and language modelling.

 

Transformer

A deep learning architecture that uses self-attention mechanisms to process sequential data, such as text or time series. Transformers power state-of-the-art models like BERT and GPT.

 

Turing Test

A test proposed by Alan Turing to evaluate a machine’s ability to exhibit intelligent behaviour indistinguishable from that of a human. Passing the Turing Test signifies advanced conversational AI capabilities.

 

Training Data

A subset of data used to train machine learning models by adjusting parameters to minimize the error on known outputs. High-quality training data is crucial for model accuracy.

 

Tree-Based Model

A machine learning model that uses decision trees as its core structure, including algorithms like Random Forests and gradient-boosted trees. These models are effective for both classification and regression tasks.

 

Tuning

The process of optimizing hyperparameters in a machine learning model to improve performance. Tuning methods include grid search, random search, and Bayesian optimization.

 

Token Embedding

A numerical representation of tokens in a vector space, capturing their semantic meaning. Pre-trained embeddings like Word2Vec and GloVe are widely used in NLP tasks.

 

Traditional AI

A term that refers to rule-based systems or models with narrower scopes of intelligence, typically used for specific tasks like chess-playing or credit scoring. It contrasts with modern AI approaches like deep learning and generative AI.

​​

​

U

​

 

Underfitting

A situation where a machine learning model is too simple to capture the underlying patterns in the data, resulting in poor performance on both training and test data. Solutions include using more complex models or adding features.

 

Unlabeled Data

Data that lacks target labels or annotations, making it suitable for unsupervised learning. Unlabeled data is often used in clustering, anomaly detection, and semi-supervised learning.

 

Unsupervised Learning

A machine learning paradigm where models are trained on unlabeled data to identify patterns, relationships, or structures. Common applications include clustering, dimensionality reduction, and generative modelling.

 

Unstructured Data

Data that does not have a predefined schema or format, such as text, images, audio, or video. Processing unstructured data often requires specialized AI techniques like natural language processing or computer vision.

​

​

V

 

​

Validation Data

A subset of data used to tune hyperparameters and evaluate a model’s performance during training. It helps prevent overfitting by providing an unbiased estimate of the model's generalization ability.

 

Vector

A one-dimensional array of numbers representing features or attributes in machine learning. Vectors are fundamental in representing data points, embeddings, and weight parameters.

 

Virtual Agent

An AI-powered entity that interacts with users through natural language or other interfaces, such as chatbots or virtual assistants. Examples include Siri, Alexa, and customer support bots.

 

Virtual Reality (VR)

A technology that creates an immersive, computer-generated environment. AI enhances VR experiences by enabling realistic interactions and adaptive content generation.

 

Video Analytics

The use of AI to extract meaningful insights and patterns from video data. Applications include surveillance, traffic monitoring, and sports analysis.

​

​

W

​​

​

Weight

A parameter in a neural network that determines the strength of the connection between neurons. During training, weights are adjusted to minimize the error and improve the model’s performance.

 

Word Embedding

A representation of words in a continuous vector space where similar words have similar representations. Examples include Word2Vec, GloVe, and FastText, which are foundational in natural language processing.

​​

​

X

 

​

X-Rank Algorithm

A ranking algorithm is often used in search engines or recommendation systems to rank items based on their relevance scores or similarity measures.

​​

​

Y

​​

​

Y-Label

The target variable in supervised learning represents the output that the model aims to predict. For example, in regression tasks, YYY is a continuous value, while in classification, it represents a class label.

 

YAML (Yet Another Markup Language)

A human-readable data serialization format often used for configuration files in AI frameworks and machine learning pipelines. YAML is lightweight and easy to edit compared to alternatives like JSON or XML.

​​

​

Z

 

​

Zero-Shot

A machine learning technique where a model can perform tasks it has not been explicitly trained on by leveraging general knowledge. For example, a language model answering questions about a topic it has never seen before.

 

Zero-Sum Game

A concept in game theory where one participant's gain or loss is exactly balanced by the losses or gains of other participants. AI models often use this concept in adversarial training, such as in Generative Adversarial Networks (GANs).

​

​

​

Top of the page​