summary.md

AI From Scratch in C++

Supervised Learning

• Linear Regression (MSE Loss, Gradient Descent)
• Logistic Regression (Sigmoid, Cross-Entropy)
• k-Nearest Neighbors (Euclidean distance, voting)
• Naive Bayes (Gaussian model, priors)
• Decision Trees (Entropy, Information Gain)
• Random Forest (Bootstrap, ensemble voting)
• Support Vector Machines (margin maximization, kernel trick)
• Perceptron (basic neural model, activation functions)
• Multilayer Perceptron (MLP, Backpropagation, ReLU/Sigmoid)
• CNN (Convolutions, Pooling, MNIST classification)
• RNN (Sequential inputs, Backpropagation Through Time)

Optimizers

• SGD (baseline optimizer)
• Momentum (accelerated SGD)
• AdaGrad (per-parameter learning rates)
• RMSProp (adaptive learning with decay)
• Adam (adaptive + momentum, most popular)

Regularization

• L1/L2 Penalties (weight shrinkage)
• Dropout (preventing co-adaptation)
• Early stopping (prevent overfitting)

Unsupervised Learning

• k-Means (clustering via distance to centers)
• PCA (dimensionality reduction via eigen decomposition)

Reinforcement Learning

• Q-Learning (value iteration, exploration vs exploitation)
• Gridworld demonstration

Evolutionary Algorithms

• Genetic Algorithm (selection, crossover, mutation)
• Fitness-based optimization

Capstone: AI Playground

Interactive SFML app with tabs/scenes:

• Linear Regression
• Optimizers (SGD, Adam, RMSProp, Momentum)
• Regularization (noisy LR vs others)
• Convolutions (blurring demo)
• Genetic Algorithm visualization
• k-Means clustering (emergent clusters)
• Q-Learning agent (gridworld)

Playground demonstrates supervised, unsupervised, RL, and evolutionary paradigms in one interface.