Last Updated: 2025-02-11

Get Started

Artificial Intelligence is shaking things up all over the place, and Asia’s bang in the middle of it with top-notch AI research, development, and innovation. In this friendly AI Glossary, you’ll find everything from the real basics to the seriously advanced—brilliant for getting a handle on AI and finding your way through this constantly expanding world.

How to Use: Feel free to hop straight to a letter in the table below or simply scroll away at your leisure. Each entry includes a handy difficulty level, a quick definition, everyday context, and links to related terms.

Table of Contents (tap a letter to jump right there)

---

~ A ~

AI (Artificial Intelligence)

Level: 🟢 Beginner
Definition: The field of computer science dedicated to creating machines or software that can perform tasks requiring human-like intelligence (e.g., problem-solving, learning).
Context: Encompasses various subfields like Machine Learning, NLP, and Robotics.
See Also: AGI, Machine Learning, Deep Learning

AGI (Artificial General Intelligence)

Level: 🔴 Advanced
Definition: A still-hypothetical form of AI with the ability to understand, learn, and apply knowledge across a broad range of tasks at a level comparable to humans.
Context: Considered the “holy grail” of AI research; raises significant ethical, societal, and philosophical questions.
See Also: AI, Ethics

Accuracy

Level: 🟢 Beginner
Definition: The proportion of correct predictions made by a model out of all predictions.
Context: Commonly used in classification; may be misleading on imbalanced datasets.
See Also: Precision, Recall

Accuracy

Activation Function

Level: 🟡 Intermediate
Definition: A mathematical function that determines the output of a neuron in a neural network, introducing non-linearity.
Context: Common types include ReLU, Sigmoid, and Tanh; essential for Deep Learning.
See Also: Neural Network, Deep Learning

Active Learning

Level: 🟡 Intermediate
Definition: An AI strategy that smartly mixes supervised and unsupervised learning, but only shouts for human help when it really needs it.
Context: Cuts down on repetitive labelling tasks by focusing human effort where it matters most, making the whole learning process a bit less of a slog.
See Also: Supervised Learning, Unsupervised Learning

Adversarial Attack

Level: 🔴 Advanced
Definition: Methods to deceive AI models (especially neural networks) by subtly altering inputs to cause incorrect outputs or classifications.
Context: Important in security-sensitive AI applications like biometric authentication.
See Also: Robustness, GAN

Algorithm

Level: 🟢 Beginner
Definition: A set of rules or steps for solving a problem or performing a computation.
Context: In AI, algorithms govern how models learn from data.
See Also: Model, Machine Learning

AutoML (Automated Machine Learning)

Level: 🟡 Intermediate
Definition: Tools and frameworks that automate the ML pipeline, including feature engineering, model selection, and hyperparameter tuning.
Context: Democratizes AI by reducing the expertise required for building effective models.
See Also: Hyperparameter Optimization, NAS

---

~ B ~

Backpropagation

Level: 🟡 Intermediate
Definition: An algorithm for training neural networks by updating weights in reverse from the output layer to the input layer.
Context: Essential to Deep Learning; works in tandem with Gradient Descent.
See Also: Gradient Descent, Neural Network

Bayesian Network

Level: 🔴 Advanced
Definition: A probabilistic model using a directed acyclic graph to represent variables and their conditional dependencies.
Context: Useful for uncertainty modeling and inference in complex domains.
See Also: Probabilistic Models

Bias

Level: 🟢 Beginner
Definition: Systematic errors in model predictions due to skewed data, flawed assumptions, or societal inequities mirrored in the training set.
Context: Can lead to discriminatory outcomes in areas like hiring, lending, or healthcare.
See Also: Fairness, Ethics

Big Data

Level: 🟢 Beginner
Definition: Extremely large and complex datasets that require specialized methods to store, process, and analyze.
Context: Underpins many AI and ML advancements by providing ample training data.
See Also: Data Mining, Data Science

---

~ C ~

Character.AI

Level: 🤖 Chatbot
Definition: A platform where you can chat with AI “characters” that have been trained to role-play as famous figures, fictional personas, or even brand-new personalities you dream up.
Context: Great for imaginative fun or storytelling—just watch out for those moments when the AI might spin you a tall tale!
Try It: Character.ai/
See Also: Prompt, Chatbot

Chatbot

Level: 🟢 Beginner
Definition: An AI-driven software application that simulates human conversation, often via text or voice interfaces.
Context: Commonly used in customer service, marketing, and information retrieval.
See Also: Conversational AI, Transformer

ChatSonic (by Writesonic)

Level: 🤖 Chatbot
Definition: A GPT-powered chatbot that helps you write blogs, emails, or social posts in a jiffy—think of it as your personal writing sidekick.
Context: Comes with real-time web browsing built in, so it can pull in updated facts and figures instead of relying on pre-trained data alone.
Try It: Writesonic Chat
See Also: Prompt, Chatbot

Claude (by Anthropic)

Level: 🤖 Chatbot
Definition: A conversational AI assistant from the folks at Anthropic—imagine ChatGPT’s clever sibling, but with a huge focus on dodging harmful or dodgy content.
Context: Delivers context-rich responses while aiming to curb “hallucinations.”
Try It: Claude AI
See Also: Prompt, Chatbot

Classification

Level: 🟢 Beginner
Definition: A supervised learning task where the model predicts a discrete category (e.g., spam vs. not spam).
Context: Measured by Accuracy, Precision, Recall, or F1-score.
See Also: Logistic Regression, Model

Clustering

Level: 🟡 Intermediate
Definition: An unsupervised learning method grouping similar data points together without pre-labeled categories.
Context: Useful in customer segmentation, anomaly detection, and data exploration.
See Also: K-Means, Unsupervised Learning

Computer Vision

Level: 🟡 Intermediate
Definition: A field of AI that enables machines to interpret and understand visual data, like images and videos.
Context: Powers facial recognition, self-driving cars, and medical image analysis.
See Also: Neural Network, YOLO

Copilot (formerly Bing Chat)

Level: 🤖 Chatbot
Definition: Microsoft’s AI-powered assistant, formerly known as Bing Chat, now built right into various Microsoft apps to offer real-time info, creative suggestions, and Q&A.
Context: Think of it as your on-the-spot AI helper—answering questions, brainstorming ideas, and even debugging code inside your favourite Microsoft products.
Try It: Copilot
See Also: Prompt, Chatbot

Conversational AI

Level: 🟡 Intermediate
Definition: AI systems designed to interact with humans via natural language, often combining speech recognition, NLP, and dialogue management.
Context: Used in virtual assistants (e.g., Alexa, Siri) and advanced customer support bots.
See Also: Chatbot, NLP

---

~ D ~

Data Augmentation

Level: 🟡 Intermediate
Definition: Techniques to increase the diversity of the training dataset (e.g., flipping images, adding noise) without collecting new data.
Context: Improves model robustness, especially in Computer Vision or Speech Recognition.
See Also: Data Mining

Level: 🖼️ Image Generation
Definition: An AI tool that conjures up images from text prompts—anything you can describe, DALL-E tries to illustrate.
Context: Great for quick concept art, quirky designs, or just having a laugh imagining “otters in outer space.”
Try It: Dall-e-3
See Also: Generative AI, Prompt

Data Mining

Level: 🟢 Beginner
Definition: The process of discovering patterns and insights in large datasets using methods at the intersection of machine learning and statistics.
Context: Often a precursor to model building in AI projects.
See Also: Big Data, Data Science

Data Science

Level: 🟢 Beginner
Definition: An interdisciplinary field that uses scientific methods, processes, and algorithms to extract knowledge from data.
Context: Data scientists often employ AI/ML techniques for predictive analytics.
See Also: Machine Learning, Big Data

Dataset

Level: 🟢 Beginner
Definition: A collection of data points—like text, images, or numbers—used to train and test AI models.
Context: Think of it as the “study material” for AI. High-quality, well-structured datasets help models learn more accurately.
See Also: Training, Data Augmentation

Deep Learning

Level: 🟡 Intermediate
Definition: A subset of ML that leverages neural networks with multiple layers to model complex patterns in data.
Context: Drives major AI breakthroughs in image recognition, NLP, and more.
See Also: Neural Network, Backpropagation

DeepSeek

Level: 🤖 Chatbot
Definition: An advanced AI search assistant that blends deep language smarts with search capabilities, aiming to dig up precise info faster than you can say “query.”
Context: Great for in-depth research tasks, though it can still slip up with made-up details now and then.
Try It: DeepSeek
See Also: Prompt, Chatbot

Differential Privacy

Level: 🔴 Advanced
Definition: A mathematical framework that adds noise to datasets or outputs to protect individual data points while allowing aggregate analysis.
Context: Critical for compliance with regulations (e.g., GDPR) and safeguarding sensitive data.
See Also: Privacy, Federated Learning

---

~ E ~

Edge AI

Level: 🟡 Intermediate
Definition: Running AI algorithms locally on edge devices (e.g., smartphones, IoT sensors) rather than centralized servers.
Context: Reduces latency, saves bandwidth, and often enhances privacy.
See Also: Edge Computing, TinyML

Edge Computing

Level: 🟢 Beginner
Definition: Processing data near the source (e.g., IoT devices) instead of sending it all to the cloud.
Context: Improves response times and data handling efficiency.
See Also: Edge AI, IoT

Ensemble Learning

Level: 🟡 Intermediate
Definition: A technique that combines multiple models to improve overall predictive performance.
Context: Methods include Bagging, Boosting, and Stacking.
See Also: Random Forest, Gradient Descent

Level: 🟢 Beginner
Definition: One complete pass through the entire training dataset by an AI model.
Context: If you go over your dataset multiple times (multiple epochs), your model might learn patterns more thoroughly—unless it overfits.
See Also: Overfitting, Training

Ethics

Level: 🟢 Beginner
Definition: Moral principles guiding AI’s design and deployment, emphasizing transparency, fairness, and accountability.
Context: Especially relevant in fields like healthcare, finance, or autonomous vehicles where AI decisions have critical impact.
See Also: Fairness, Bias

---

~ F ~

Fairness

Level: 🟢 Beginner
Definition: Ensuring AI outcomes do not discriminate or harm certain groups, especially vulnerable or historically disadvantaged populations.
Context: Closely tied to bias mitigation and ethical AI guidelines.
See Also: Bias, Ethics

Federated Learning

Level: 🔴 Advanced
Definition: A decentralized ML approach where devices train local models and share only aggregated updates, not raw data.
Context: Maintains data privacy and is suitable for large-scale, distributed systems (smartphones, IoT).
See Also: Edge AI, Differential Privacy

Feature Engineering

Level: 🟡 Intermediate
Definition: The process of selecting, creating, or transforming input variables (features) to improve model performance.
Context: Vital in traditional ML workflows; neural networks often learn features automatically.
See Also: Model

Few-Shot Learning

Level: 🔴 Advanced
Definition: Techniques enabling models to generalize from only a small number of labeled examples.
Context: Crucial in scenarios where data collection is expensive or limited.
See Also: Zero-Shot Learning, Meta-Learning

Fine-Tuning

Level: 🟡 Intermediate
Definition: Taking a model that’s already been trained on a broad dataset and giving it extra training on more specific data, so it gets really good at a narrower task.
Context: Common with large language models—why start from scratch when you can build on existing knowledge?
See Also: Transfer Learning, LLM

F1-Score

Level: 🟡 Intermediate
Definition: The harmonic mean of precision and recall, offering a single metric when dealing with class imbalance.
Context: Often used in classification to balance false positives and false negatives.
See Also: Precision, Recall

---

~ G ~

GAN (Generative Adversarial Network)

Level: 🔴 Advanced
Definition: A neural network framework with two models—Generator and Discriminator—competing to create realistic data.
Context: Generates synthetic images, text, or other data; used in art, game design, and data augmentation.
See Also: Adversarial Attack, Unsupervised Learning

Gemini (by Google)

Level: 🤖 Chatbot
Definition: Google’s next-generation AI model, taking over where Bard left off—like your genius mate who’s absorbed Google’s entire knowledge base and can chat on any topic.
Context: Touted as a major leap forward in conversational AI, backed by Google’s vast machine learning resources.
Try It: Google Gemini
See Also: Prompt, Chatbot

Generative AI

Generative AI

Level: 🟡 Intermediate
Definition: AI that doesn’t just analyse data, but actually creates brand-new content—like text, images, and even music.
Context: Powering everything from AI art apps to chatbots, it’s shaking up industries with fresh ways to produce, well, just about anything.
See Also: GAN, LLM, Transformer

GPT (Generative Pre-Trained Transformer)

Level: 🔴 Advanced
Definition: A type of large language model (LLM) that harnesses the Transformer architecture to generate remarkably coherent text—famously exemplified by OpenAI’s GPT series.
Context: Known for its massive scale and wide-ranging capabilities, from drafting emails to crafting code. GPT models underscore just how far AI can go in churning out human-like content.
See Also: Large Language Model (LLM), Transformer

GPU (Graphics Processing Unit)

Level: 🟢 Beginner
Definition: A specialised chip originally made for rendering graphics, but brilliant at handling the parallel computations needed for AI tasks.
Context: Speeds up training large models—basically, the horsepower behind modern AI breakthroughs.
See Also: Edge AI, Deep Learning

Gradient Descent

Level: 🟡 Intermediate
Definition: An optimization algorithm that adjusts model parameters to minimize a cost (loss) function by following the gradient.
Context: Integral to training most neural networks, with variants like Stochastic and Mini-Batch.
See Also: Backpropagation, Loss Function

Grok (by xAI)

Level: 🤖 Chatbot
Definition: A conversational assistant from Elon Musk’s xAI venture, aiming to bring a fresh twist to the AI chatbot scene—still in its early days but garnering heaps of attention.
Context: Positioned as a direct rival to heavyweights like ChatGPT, with an eye on open, unfiltered conversations.
Try It: X ai
See Also: Prompt, Chatbot

---

~ H ~

Hallucination

Level: 🟡 Intermediate
Definition: When an AI system churns out information that sounds perfectly credible but is actually made-up—essentially presenting fiction as fact.
Context: Often pops up in large language models, highlighting the need for verification and fact-checking, especially in high-stakes environments.
See Also: XAI, Bias

Heuristic

Level: 🟢 Beginner
Definition: A rule-of-thumb or approximate method for quickly finding a solution when exact algorithms are impractical.
Context: Commonly used in AI search problems (e.g., pathfinding, game AI).
See Also: Algorithm

HuggingChat (by Hugging Face)

Level: 🤖 Chatbot
Definition: An open-source chatbot platform run by Hugging Face, letting you converse with cutting-edge AI models without the usual paywalls or sign-ups.
Context: Ideal for users who want a free, community-driven environment to experiment with AI text generation.
Try It: Hugging Face Chat
See Also: Prompt, Chatbot

Hyperparameter

Level: 🟡 Intermediate
Definition: A parameter set before training (e.g., learning rate, number of layers) that controls the training process.
Context: Fine-tuning hyperparameters is crucial for model performance.
See Also: Hyperparameter Optimization

Hyperparameter Optimization

Level: 🔴 Advanced
Definition: Techniques (like grid search, random search, Bayesian optimization) for finding the best hyperparameter configurations.
Context: Can drastically improve model accuracy and speed.
See Also: AutoML, Gradient Descent

---

~ I ~

Ideogram

Level: 🖼️ Image Generation
Definition: A text-to-image generator that claims to handle even the trickiest text-based prompts, focusing on clearer typography in AI-generated designs.
Context: Useful if you need stylised words and graphics in your AI images, rather than just abstract scenes.
Try It: https://ideogram.ai/
See Also: Generative AI, Prompt

Inference

Level: 🟢 Beginner
Definition: The process of using a trained model to make predictions on new, unseen data.
Context: Often optimized separately from training for speed, especially in production.
See Also: Model

IoT (Internet of Things)

Level: 🟢 Beginner
Definition: A network of connected devices (e.g., sensors, appliances) that collect and exchange data.
Context: AI at the edge (Edge AI) can process IoT data in real time for applications like smart homes, industrial monitoring.
See Also: Edge Computing

---

~ J ~

Jupyter Notebook

Level: 🟢 Beginner
Definition: An open-source web tool that allows you to combine live code, visualizations, and narrative text in a single document.
Context: Widely used in data science and AI for exploratory analysis and prototyping.
See Also: Python, Data Science

---

~ K ~

K-Means

Level: 🟢 Beginner
Definition: A popular clustering algorithm partitioning data into K distinct clusters by minimizing within-cluster variance.
Context: Often a go-to algorithm for quick segmentation tasks.
See Also: Clustering, Unsupervised Learning

Knowledge Graph

Level: 🟡 Intermediate
Definition: A structure of entities (nodes) and their relationships (edges) representing real-world facts.
Context: Used by search engines (e.g., Google) for better data connectivity.
See Also: Ontology, Semantic Web

---

~ L ~

Large Language Model (LLM)

Level: 🟡 Intermediate
Definition: A neural network (often with billions of parameters) trained on massive text corpora to understand and generate language.
Context: Powers advanced NLP tasks, including chatbots, summarization, and translation.
See Also: Transformer

Learning Rate

Level: 🟡 Intermediate
Definition: A hyperparameter controlling how much model weights update during each training step.
Context: Too large can cause divergence; too small can slow learning.
See Also: Gradient Descent

Logistic Regression

Level: 🟢 Beginner
Definition: A statistical model often used for binary classification (e.g., predicting yes/no outcomes).
Context: Known for its simplicity and interpretability; often a baseline for more complex models.
See Also: Classification

Loss Function

Level: 🟡 Intermediate
Definition: A measure of how far a model’s predictions deviate from actual targets.
Context: Guides the learning process via Gradient Descent.
See Also: Training

---

~ M ~

Machine Learning (ML)

Level: 🟢 Beginner
Definition: A subset of AI that allows systems to learn from data and make predictions or decisions with minimal human intervention.
Context: Encompasses supervised, unsupervised, and reinforcement learning.
See Also: Deep Learning, Data Science

Midjourney

Level: 🖼️ Image Generation
Definition: A Discord-based AI art generator famous for producing jaw-dropping, highly detailed images—just type in your prompt, and watch the magic happen.
Context: Hugely popular among digital artists and casual doodlers alike; known for an imaginative, painterly aesthetic.
Try It: https://midjourney.com/
See Also: Generative AI, Prompt

Meta AI

Level: 🤖 Chatbot
Definition: Meta’s (formerly Facebook) AI chatbot, geared towards a broad user base across Messenger, Instagram, and more—like a digital mate that can help with queries or just have a natter.
Context: Part of Meta’s larger ambition to weave AI assistants into everyday social interactions, not just standalone apps.
Try It: Meta AI

Meta-Learning

Level: 🔴 Advanced
Definition: “Learning to learn,” where a model leverages knowledge from previous tasks to learn new tasks more efficiently.
Context: Useful in Few-Shot or Zero-Shot Learning scenarios.
See Also: Few-Shot Learning, Transfer Learning

Mistral

Level: 🤖 Chatbot
Definition: A newly developed AI model focussed on open-source collaboration, designed to deliver top-notch results without ballooning into a massive, unwieldy giant.
Context: Perfect for those who like to tinker—ideal if you want to tweak or fine-tune a chatbot under the hood.
Try It: Le Chat
See Also: Prompt, Chatbot

Model

Level: 🟢 Beginner
Definition: A mathematical representation of a real-world process, trained from data to perform specific tasks (e.g., classification).
Context: Types include neural networks, decision trees, and linear models.
See Also: Training, Inference

Multi-Task Learning

Level: 🔴 Advanced
Definition: A learning paradigm in which a single model is trained to handle multiple tasks simultaneously.
Context: Can improve performance by leveraging shared representations across tasks.
See Also: Transfer Learning

My AI (by Snapchat)

Level: 🤖 Chatbot
Definition: Snapchat’s built-in AI buddy, ready to chat about your day, suggest filters, or even join group convos for a bit of fun banter.
Context: It’s integrated right into Snapchat, so no separate app needed—handy for on-the-fly chats and quick tips.
Try It: Snapchat My AI
See Also: Prompt, Chatbot

---

~ N ~

NAS (Neural Architecture Search)

Level: 🔴 Advanced
Definition: Automated methods to discover optimal neural network architectures, often outperforming manually designed networks.
Context: Very computationally intensive but can yield state-of-the-art results.
See Also: AutoML, Hyperparameter Optimization

Natural Language Processing (NLP)

Level: 🟢 Beginner
Definition: A branch of AI focusing on the interaction between computers and human language.
Context: Tasks include sentiment analysis, machine translation, and text summarization.
See Also: Transformer, Tokenization

Neural Network

Level: 🟡 Intermediate
Definition: A computational model inspired by biological neurons, consisting of layers that learn hierarchical representations of data.
Context: The core of Deep Learning.
See Also: Deep Learning, Activation Function

NLP Pipeline

Level: 🟢 Beginner
Definition: The sequence of steps (e.g., tokenization, part-of-speech tagging) applied to text data in NLP.
Context: Ensures consistent processing of language data prior to modeling.
See Also: NLP

---

~ O ~

Ontology

Level: 🟡 Intermediate
Definition: A formal structure representing concepts within a domain and relationships between them.
Context: Used in semantic web, knowledge graphs, and advanced AI reasoning.
See Also: Knowledge Graph, Semantic Web

Overfitting

Level: 🟢 Beginner
Definition: When a model learns the training data’s noise or details too well, harming generalization to new data.
Context: Common solutions include regularization, cross-validation, or early stopping.
See Also: Underfitting, Regularization

---

~ P ~

Pi (by Inflection AI)

Level: 🤖 Chatbot
Definition: A personal AI companion from Inflection AI that aims to offer a friendly, chat-based approach to everything from scheduling reminders to life advice.
Context: Designed for one-on-one help, focusing on emotional understanding as well as factual answers—sort of a cross between a notepad and a confidant.
Try It: https://heypi.com/
See Also: Prompt, Chatbot

Poe (by Quora)

Level: 🤖 Chatbot
Definition: A chatbot aggregator where you can switch between multiple AI models (including GPT variants) in one place—like a sampler platter of AI engines.
Context: Perfect if you’re curious to test the quirks of different chatbots without hopping between loads of sites.
Try It: https://poe.com/
See Also: Prompt, Chatbot

Precision

Level: 🟡 Intermediate
Definition: The fraction of predicted positives that are truly positive.
Context: Important when false positives are costly (e.g., medical diagnoses).
See Also: Recall, F1-Score

Privacy

Level: 🟢 Beginner
Definition: Protecting personal or sensitive data from unauthorized access or disclosure, often through techniques like encryption or differential privacy.
Context: Critical in healthcare, finance, and any application handling sensitive user data.
See Also: Differential Privacy, Federated Learning

Probabilistic Models

Level: 🟡 Intermediate
Definition: Models that incorporate probability distributions to handle uncertainty (e.g., Bayesian networks).
Context: Useful in domains like finance, weather forecasting, and medical diagnostics where data is noisy or incomplete.
See Also: Bayesian Network

Prompt

Level: 🟢 Beginner
Definition: A piece of text (or sometimes an image or other input) given to an AI model to guide its response or output.
Context: In chatbots like ChatGPT, the prompt is essentially your question or statement—framing it clearly can help the AI produce a more accurate or creative answer.
See Also: Prompt Engineering, Generative AI

Prompt Engineering

Level: 🟡 Intermediate
Definition: Crafting clever inputs or questions so an AI model (especially a language model) gives you the most useful answers.
Context: Hugely important in conversational AI—getting the prompt right can be the difference between a spot-on reply and a confusing jumble.
See Also: GPT, Generative AI

Python

Level: 🟢 Beginner
Definition: A high-level programming language widely used in AI/ML due to its readability and extensive libraries (NumPy, Pandas, PyTorch, TensorFlow).
Context: De facto standard for AI prototyping and production.
See Also: Jupyter Notebook

---

~ Q ~

Q-Learning

Level: 🔴 Advanced
Definition: A reinforcement learning technique that learns a quality function (Q) for actions in given states.
Context: Popular in game AI and robotics for sequential decision-making tasks.
See Also: Reinforcement Learning

Quantum Computing

Level: 🔴 Advanced
Definition: Computing based on quantum-mechanical phenomena, enabling potentially vast speedups for certain algorithms.
Context: Could revolutionize AI for optimization, cryptography, and complex simulations.
See Also: Algorithm

---

~ R ~

Random Forest

Level: 🟡 Intermediate
Definition: An ensemble method combining many decision trees to improve prediction accuracy and reduce overfitting.
Context: A robust, commonly used baseline model.
See Also: Ensemble Learning

Recall

Level: 🟡 Intermediate
Definition: The fraction of actual positives that are correctly identified by the model.
Context: Crucial when missing positives is costly (e.g., disease detection).
See Also: Precision, F1-Score

Recommender System

Level: 🟡 Intermediate
Definition: Algorithms that predict user preferences to suggest relevant items (e.g., movies, products).
Context: Widely used in e-commerce, streaming services, and social media.
See Also: Machine Learning

Regularization

Level: 🟡 Intermediate
Definition: Techniques (L1, L2, Dropout) to prevent overfitting by penalizing complex models.
Context: Helps models generalize better to unseen data.
See Also: Overfitting, Loss Function

Reinforcement Learning

Level: 🔴 Advanced
Definition: A learning paradigm where an agent learns to perform actions in an environment to maximize a cumulative reward.
Context: Used in robotics, gaming (AlphaGo), and self-driving vehicles.
See Also: Q-Learning

Replika

Level: 🤖 Chatbot
Definition: A virtual companion app where you can chat and bond with an AI “friend” that learns your interests, personality, and conversation style over time.
Context: Emphasises emotional support and personal connection, more than just fetching facts or drafting content.
Try It: Replika
See Also: Prompt, Chatbot

Robustness

Level: 🟡 Intermediate
Definition: The ability of an AI model to maintain performance under adversarial inputs or variations in data.
Context: Critical for safety-critical applications like autonomous driving or healthcare.
See Also: Adversarial Attack

---

~ S ~

Semantic Web

Level: 🟡 Intermediate
Definition: A framework for sharing data across the web in a machine-readable format, enabling better data interoperability.
Context: Facilitates AI systems in understanding and linking information.
See Also: Ontology, Knowledge Graph

Stable Diffusion

Level: 🖼️ Image Generation
Definition: An open-source image model that translates text prompts into high-quality visuals, all without relying on proprietary code.
Context: Perfect for DIY enthusiasts who want to run AI art on their own hardware, or customise the model under the hood.
Try It: Stability AI
See Also: Generative AI, Prompt

Supervised Learning

Level: 🟢 Beginner
Definition: A type of ML where the model learns from labeled data (input-output pairs).
Context: Includes tasks like classification and regression.
See Also: Unsupervised Learning, Reinforcement Learning

Support Vector Machine (SVM)

Level: 🟡 Intermediate
Definition: A supervised learning algorithm finding the optimal hyperplane to separate different classes in high-dimensional space.
Context: Effective for medium-sized, well-structured datasets.
See Also: Classification

~ T ~

Test Set

Level: 🟢 Beginner
Definition: A final set of data used to evaluate the performance of a fully trained model.
Context: Essential for unbiased performance assessment.
See Also: Validation Set, Training

TinyML

Level: 🔴 Advanced
Definition: Deploying ML models on ultra-low-power microcontrollers, enabling AI on embedded devices.
Context: Useful for battery-powered IoT sensors, wearables, and remote monitoring.
See Also: Edge AI, Federated Learning

Tokenization

Level: 🟢 Beginner
Definition: Splitting text into smaller units (tokens), such as words or subwords, for NLP tasks.
Context: A foundational step in any text processing pipeline.
See Also: NLP Pipeline

Training

Level: 🟢 Beginner
Definition: The process of feeding data into a model and adjusting parameters to minimize a loss function.
Context: Can be computationally intensive, especially for deep networks.
See Also: Inference, Loss Function

Transfer Learning

Level: 🟡 Intermediate
Definition: Using knowledge gained from one task to improve learning on another, related task.
Context: Common in computer vision (pretrained ImageNet models) and NLP (pretrained Transformers).
See Also: Multi-Task Learning, Few-Shot Learning

Transformer

Level: 🔴 Advanced
Definition: A neural network architecture that uses self-attention mechanisms and excels in NLP tasks.
Context: Forms the basis of many state-of-the-art language models (e.g., BERT, GPT).
See Also: LLM, NLP

~ U ~

Underfitting

Level: 🟢 Beginner
Definition: When a model fails to capture underlying trends in the data, leading to low accuracy even on training data.
Context: Can be addressed by using more complex models or training longer.
See Also: Overfitting, Regularization

Unsupervised Learning

Level: 🟢 Beginner
Definition: A learning approach without labeled data, where the model aims to find hidden patterns or groupings.
Context: Clustering and dimensionality reduction are popular unsupervised methods.
See Also: Clustering, GAN

---

~ V ~

Validation Set

Level: 🟢 Beginner
Definition: A subset of data used during model training to tune hyperparameters and avoid overfitting.
Context: Helps identify optimal model settings before final testing.
See Also: Test Set, Training

Vector Embedding

Level: 🟡 Intermediate
Definition: A numeric representation of data (words, images) in a lower-dimensional space, capturing semantic or structural similarities.
Context: Commonly used in NLP (Word2Vec, GloVe) and recommender systems.
See Also: Tokenization, NLP

---

~ W ~

Weight

Level: 🟡 Intermediate
Definition: A parameter in a neural network that transforms input data within each neuron’s computation.
Context: Adjusted during backpropagation to minimize loss.
See Also: Backpropagation, Neural Network

Word2Vec

Level: 🟡 Intermediate
Definition: A method for learning vector embeddings of words, capturing semantic and syntactic similarities.
Context: Pioneered large-scale adoption of embedding-based NLP approaches.
See Also: Vector Embedding, NLP

---

~ X ~

XAI (Explainable AI)

Level: 🔴 Advanced
Definition: Techniques and frameworks aimed at making AI models more transparent and interpretable.
Context: Crucial in regulated industries (finance, healthcare) and for building public trust in AI systems.
See Also: Ethics, Bias

---

~ Y ~

YOLO (You Only Look Once)

Level: 🟡 Intermediate
Definition: A real-time object detection algorithm known for speed and accuracy in identifying objects in images.
Context: Widely applied in surveillance, autonomous driving, and robotics.
See Also: Computer Vision, Deep Learning

---

~ Z ~

Zero-Shot Learning

Level: 🔴 Advanced
Definition: A technique enabling models to classify or perform tasks on categories they’ve never been explicitly trained on.
Context: Relies on semantic knowledge transfer; valuable in rapidly changing domains.
See Also: Few-Shot Learning, Transfer Learning

---

Bookmark This Page

This AI Glossary is designed to be a living resource, adaptable as new technologies and methodologies emerge—especially in the vibrant AI ecosystems across Asia. If you notice a missing term or have an updated definition, let us know in the comments or via our form. Bookmark this page and check back for regular updates!

You may also like: