Interactive Learning

AI Explorer

Tap any concept to explore โ€” with real-world scenes and diagrams that make AI click.

Start Here
๐Ÿค–

Artificial Intelligence

Imagine teaching a dog new tricks โ€” but instead of a dog, it's a computer, and instead of treats, you use data. AI is the science of making machines smart enough to do things that normally require human thinking: recognising faces, understanding speech, recommending shows, diagnosing diseases, or even driving cars.

Hey Alexa! Turn off the lights and play jazz ๐ŸŽต ๐Ÿ’ก ๐Ÿค– AI-powered home
AI in your home โ€” smart speakers understand voice commands using NLP, control devices, and learn your habits
Artificial Intelligence โš™๏ธ Machine Learning ๐Ÿง  Deep Learning ๐Ÿ’ฌ NLP Language ๐Ÿ‘๏ธ Computer Vision ๐ŸŽฎ Reinforce- ment Learning
The five major branches of AI โ€” each powering different real-world capabilities

๐Ÿ‘‡ Explore each branch below

Major Branches of AI
โš™๏ธ

Machine Learning

ML is the backbone of modern AI. Instead of writing rules by hand, you show the machine thousands of examples and it figures out the patterns itself โ€” like a child who learns to identify dogs by seeing enough photos.

๐Ÿ“ง Inbox 9 โœ“ Meeting at 3pm today Sarah โ€” Hi, just confirming our meeting... INBOX โœ— YOU WON $1,000,000!!! Click here to claim your prize now... SPAM โœ“ Your order has shipped ๐Ÿ“ฆ Amazon โ€” Your package is on the way... INBOX โš™๏ธ ML classifies each email automatically
Spam detection โ€” ML learns from millions of emails to classify new ones as spam or legit
๐ŸŽฌ StreamAI โš™๏ธ ML RECOMMENDS FOR YOU ๐Ÿš€ Space Odyssey ๐Ÿค– AI Rising ๐ŸŒŒ Galaxy Quest ๐Ÿ”ฌ Interstellar โ˜…โ˜…โ˜…โ˜…โ˜† โ˜…โ˜…โ˜…โ˜…โ˜… โ˜…โ˜…โ˜…โ˜…โ˜† โ˜…โ˜…โ˜…โ˜…โ˜… Based on your watch history ยท 94% match
Movie recommendations โ€” ML analyses your viewing patterns to suggest shows you'll love
โš™๏ธ Machine Learning ๐Ÿท๏ธ Supervised labelled data โ†’ predict ๐Ÿ“ˆ Regression ๐Ÿ—‚๏ธ Classification ๐Ÿ” Unsupervised no labels โ†’ find patterns ๐Ÿซง Clustering ๐Ÿ“‰ Dim. Reduction ๐Ÿ”€ Semi-Supervised mix of both approaches

Three ML paradigms and their key sub-types

Types of Machine Learning

๐Ÿท๏ธ Supervised Learning

The model trains on labelled examples with correct answers โ€” like a student learning from an answer key. Two core tasks: Regression (predict a number) and Classification (predict a category).

๐Ÿ“ˆ Regression: House Price Predictor House Size (sqft) Price ($) ML line Predicts: if size = 2000sqft โ†’ price = $320,000
Regression โ€” ML draws the best-fit line to predict house prices from size
๐Ÿ—‚๏ธ Classification: Disease Detector Brain MRI Scan โ†’ ML Model Diagnosis โš ๏ธ Tumour Detected Confidence: 94.2% โ†’ Refer to specialist Trained on 50,000 labelled scans
Classification โ€” ML trained on labelled scans predicts whether a tumour is present
๐Ÿ“ˆ
Regression
Predicts a number. House prices, temperature, stock values.
๐Ÿ—‚๏ธ
Classification
Predicts a category. Spam/Not Spam, Cat/Dog, Healthy/Disease.
๐ŸŒณ
Decision Trees
Splits data like a flowchart โ€” each branch is a decision rule.
๐Ÿ“Š
SVM
Draws the best boundary between classes in feature space.
๐Ÿค
K-Nearest Neighbours
Classifies based on majority label among its closest neighbours.
๐ŸŒฒ
Random Forest
Hundreds of decision trees voting together for accuracy.

๐Ÿ” Unsupervised Learning

No labels, no teacher. The model explores raw data and discovers hidden structure on its own โ€” like grouping customers by shopping habits without anyone telling it which group they belong to.

๐Ÿซง Customer Segmentation Raw data (no labels) โ†’ K-Means Discovered groups Budget shoppers Premium buyers Occasional buyers
Clustering โ€” discovers natural customer groups without any labels, to target marketing
๐Ÿšจ Anomaly Detection โ€” Bank Fraud Recent Transactions โœ“ Coffee Shop โ€” $4.50 09:14 โœ“ Grocery Store โ€” $67.20 11:30 โœ“ Gas Station โ€” $42.00 13:45 โš ๏ธ FLAGGED โ€” Electronics Store Nigeria โ€” $2,840 15:02 Location anomaly ยท Unusual amount ยท Pattern mismatch โœ“ Pharmacy โ€” $18.00 16:20 ๐Ÿ” ML detected anomaly โ€” card blocked instantly
Anomaly detection โ€” unsupervised ML learns normal spending patterns and flags unusual transactions
๐Ÿซง
Clustering
Groups similar data. K-Means, DBSCAN, Hierarchical.
๐Ÿ“‰
Dimensionality Reduction
Compresses data while keeping meaning. PCA, t-SNE, UMAP.
๐Ÿ”—
Association Rules
Finds co-occurring items. "Customers who buy X also buy Y."
๐Ÿ—๏ธ
Autoencoders
Neural nets that compress then reconstruct โ€” learn representations.

๐Ÿ”€ Semi-Supervised Learning

A practical middle ground โ€” a small labelled set + a large unlabelled set. Common when labelling everything is expensive, like annotating millions of medical scans.

๐Ÿฅ Medical Scan Labelling Expert Labelled (100) ๐Ÿซ โœ“ Normal ๐Ÿซ โš  Tumour + 98 more labelled scans Cost: ๐Ÿ’ฐ๐Ÿ’ฐ๐Ÿ’ฐ (very expensive) Unlabelled (10,000) ๐Ÿซ ? ๐Ÿซ ? ๐Ÿซ ? + 9,997 moreโ€ฆ Cost: ๐Ÿ’š (almost free) โœ… Better Model 95% accuracy vs 78% with labelled-only
Semi-supervised: a handful of expert-labelled scans + thousands of unlabelled ones = powerful AI model
๐Ÿ”€ How Semi-Supervised Works 1๏ธโƒฃ Train on labelled small expert dataset 2๏ธโƒฃ Pseudo-label predict unlabelled data 3๏ธโƒฃ Retrain with all labelled + pseudo-labelled 4๏ธโƒฃ Better model โœ“ higher accuracy achieved Common Use Cases ๐Ÿฅ Medical Imaging ๐ŸŒ Web Classification ๐Ÿ”ค Text Tagging few labelled scans tag a few pages few labelled docs
The pseudo-labelling cycle โ€” iteratively improves the model using its own predictions
๐Ÿง 

Deep Learning

Deep Learning uses layered Neural Networks inspired by the brain. The "deep" refers to many layers stacked together โ€” each one learns increasingly abstract features: from raw pixels to edges, shapes, objects, and finally meaning.

๐Ÿ–ผ๏ธ Image Recognition Pipeline Input Photo Layer 1 Edges Layer 2 Shapes Layer 3 Features Output ๐Ÿ• DOG 98.7% Neural Network โ€” Each neuron learns a weight Dog Architectures: CNN โ€” images RNN โ€” sequences Transformer โ€” text
Deep learning pipeline โ€” raw photo passes through layers, each extracting richer features, ending in "Dog 98.7%"
โšก Transformers โ€” Powering ChatGPT & Claude Input sentence (tokenised) Hey Siri I am hungry ! Attention: "Siri" strongly attends to "hungry" ๐ŸŽฏ ๐Ÿ–ผ๏ธ CNN Convolutional slides a filter over the image โ†’ photos, video ๐Ÿ”„ RNN/LSTM Recurrent โ€” has memory of past tokens/steps โ†’ speech, text โšก Transformer Attention-based, parallel, powerful GPT, Claude, BERT โ†’ LLMs, NLP
Transformers use "attention" โ€” each word looks at all other words to understand context
Neural Network Architectures
๐Ÿ”—
ANN
Artificial Neural Network โ€” the basic foundation: input, hidden, output layers.
๐Ÿ–ผ๏ธ
CNN
Convolutional NN โ€” king of image recognition and computer vision.
๐Ÿ”„
RNN
Recurrent NN โ€” handles sequences; great for time-series and speech.
๐Ÿงฌ
LSTM
Long Short-Term Memory โ€” RNN with memory gates for long sequences.
โšก
Transformer
Attention-based architecture powering GPT, Claude, BERT, Gemini.
๐ŸŽจ
GAN
Two networks compete โ€” generator vs discriminator. Creates realistic images.
๐Ÿ’ฌ

Natural Language Processing

NLP allows computers to read, write, and understand human language. Every time you say "Hey Siri, I am hungry" โ€” NLP parses your words, understands intent, and triggers a response. It bridges messy, ambiguous human language and precise machine logic.

Hey Siri, I am hungry! Find me nearby restaurants ๐Ÿ• NLP Pipeline: "Hey Siri, I am hungry" Tokenise โ†’ Intent = food_search โ†’ Entities = [hungry] โ†’ Action = find_restaurants
"Hey Siri, I am hungry!" โ€” NLP detects intent (food search), extracts entities, and triggers restaurant finder
๐Ÿ’ฌ NLP Tasks in the Real World ๐Ÿ˜Š Sentiment Analysis "This pizza was amazing!" โ†’ Positive (96%) ๐ŸŒ Machine Translation "Bonjour, comment รงa va?" โ†’ "Hello, how are you?" ๐Ÿท๏ธ Named Entity Recognition "Elon Musk founded SpaceX in California in 2002" โ†’ Person ยท Org ยท Location ยท Date ๐Ÿค– Chatbot / LLM User: "Write me a poem" โ†’ GPT, Claude, Gemini generate human-like text NLP Pipeline Tokenise Vectorise Model (LLM) Intent/Entity Response
NLP powers sentiment analysis, translation, entity recognition, chatbots โ€” any task involving language
๐Ÿ˜Š
Sentiment Analysis
Detects if text is positive, negative, or neutral. Used in reviews and social media.
๐Ÿ“
Text Summarisation
Condenses long documents into short key summaries automatically.
๐ŸŒ
Machine Translation
Converts text between languages. Google Translate uses NLP.
๐Ÿค–
Chatbots / LLMs
Conversational AI โ€” GPT, Claude, and Gemini are all NLP-powered.
๐Ÿท๏ธ
Named Entity Recognition
Identifies names, places, dates, and organisations in text.
๐Ÿ‘๏ธ

Computer Vision

Computer Vision gives machines the ability to see and interpret visual information โ€” photos, videos, live camera feeds. From unlocking your phone with your face to self-driving cars reading road signs, CV is everywhere.

๐Ÿš— Self-Driving Car โ€” Computer Vision Car โ€” 97% Person STOP Sign ๐Ÿš— Objects: 2 Car: 97% ๐Ÿšถ Persons: 1 84% confidence ๐Ÿ›‘ Signs: 1 STOP โ€” 99%
Self-driving CV โ€” detects cars, pedestrians, lane lines and road signs in real time to steer safely
๐Ÿฅ Medical Imaging โ€” CV Diagnosis โš  Nodule 91% AI Diagnosis Scan Analysis Complete โš ๏ธ Anomaly Detected Left upper lobe Confidence: 91.4% Size: ~8mm nodule โ†’ Recommend CT scan CNN trained on 200,000 labelled medical scans 94% accuracy โ€” assists radiologists to detect early-stage disease
Medical CV โ€” CNN analyses chest X-rays and flags anomalies, helping radiologists detect disease earlier
๐Ÿ”Ž
Image Classification
Labels an entire image โ€” "this is a cat." One label per image.
๐Ÿ“ฆ
Object Detection
Boxes multiple objects in an image. YOLO, Faster R-CNN.
๐ŸŽญ
Image Segmentation
Labels every single pixel in an image by object type.
๐Ÿ˜€
Face Recognition
Identifies or verifies a person from their face geometry.
๐ŸŽฌ
Video Analysis
Tracks motion, actions, and events across video frames.
๐Ÿš—
Autonomous Driving
Reads lanes, signs, pedestrians and other cars in real time.
๐ŸŽฎ

Reinforcement Learning

RL is inspired by how animals learn through rewards and punishments. An agent takes actions in an environment, receives rewards for good decisions and penalties for bad ones, and gradually learns the best strategy โ€” without any human-provided examples.

๐ŸŽฎ RL Training a Game AI $ $ $ +10 pts +50 pts -50 pts Score: 1,240 Episode: 10,482 Best: 24,800 Avg Q: 0.82 RL agent plays 10,000+ episodes โ€” learns: collect coins, avoid spikes No rules given โ€” discovers strategy purely through reward signals
Game AI learns from thousands of self-play episodes โ€” reward for coins, penalty for spikes
๐Ÿ”„ The Reinforcement Learning Loop ๐Ÿค– Agent has a policy (strategy) picks actions to maximise reward ๐ŸŒ Environment game, robot world real or simulated Action +Reward / โˆ’Penalty New State (observation) ๐Ÿ† Goal Maximise total reward over time Real-World Applications โ™Ÿ๏ธ AlphaGo Chess & Go champion ๐Ÿค– Robotics locomotion & grasping ๐Ÿ’น Trading Bots market strategy ๐Ÿ—ฃ๏ธ RLHF ChatGPT tuning
The RL loop: Agent acts โ†’ Environment responds with reward/penalty โ†’ Agent improves policy over thousands of trials
๐Ÿ•น๏ธ
Q-Learning
Learns a Q-table of values โ€” best action for each state.
๐Ÿ•ธ๏ธ
Deep Q-Network (DQN)
Neural network approximates Q-values for complex environments.
๐Ÿ†
Policy Gradient
Directly optimises the action strategy the agent uses.
๐Ÿค–
Actor-Critic
Combines policy gradient (actor) with value estimation (critic).
๐ŸŒ
Real-World Uses
AlphaGo, robotics, trading bots, recommendations, RLHF in LLMs.