From first principles to modern architectures. Every algorithm animated, every concept interactive. Built on the latest research papers — from gradient descent to diffusion models and transformers.
Carefully structured from mathematical foundations to cutting-edge architectures
Gradient Descent, SGD, Momentum, Adam — the engines behind all learning algorithms
K-Means, PCA, t-SNE, UMAP — finding structure in high-dimensional data
Linear Regression, Logistic Regression, SVMs and the geometry of decision boundaries
Biological inspiration to universal approximation — forward pass, backpropagation, and activations
CNNs, RNNs, the Attention mechanism, and the Transformer architecture powering GPT, BERT and beyond
GANs, VAEs, Diffusion Models (DDPM/DDIM), and Stable Diffusion — the state of generative AI today
Every algorithm covered, with complexity and use-case
| Algorithm | Chapter | Type | Time Complexity | Key Use Case |
|---|---|---|---|---|
| Gradient Descent | 1 | Optimization | O(n·d·iter) | Parameter optimization in all models |
| Adam Optimizer | 1 | Optimization | O(n·d·iter) | Training deep networks faster |
| K-Means | 2 | Unsupervised | O(n·k·iter) | Customer segmentation, data compression |
| PCA | 2 | Unsupervised | O(n·d²) | Dimensionality reduction, noise removal |
| Linear Regression | 3 | Supervised | O(n·d²) | Price prediction, trend analysis |
| Logistic Regression | 3 | Supervised | O(n·d·iter) | Binary classification baselines |
| SVM | 3 | Supervised | O(n²·d) | High-dimensional classification |
| Backpropagation | 4 | Deep Learning | O(L·n·d²) | Training neural networks |
| Self-Attention | 5 | Deep Learning | O(n²·d) | Language and vision transformers |
| Transformer | 5 | Deep Learning | O(n²·d) | GPT, BERT, ViT architectures |
| VAE | 6 | Generative | O(L·n·d) | Image synthesis, latent space learning |
| Diffusion (DDPM) | 6 | Generative | O(T·n·d) | Image/audio generation (Stable Diffusion) |
This is an enhanced interactive version of the ml-visualized concept, rebuilt with live JavaScript visualizations instead of static GIFs, incorporating the latest research continuously.
Each chapter derives algorithms mathematically, then animates them step-by-step. You can interact with parameters, pause animations, and see how changes affect the learning process in real time.