Learn ML from Scratch: Beginner to Advanced

A rigorous, step-by-step interactive curriculum. Master foundational mathematical derivations, vector calculus, pure NumPy/PyTorch implementations from scratch, and modern 2026 reasoning LLM architectures.

🎓 Track Progress: 0 Completed
0%

Choose where to start

New to machine learning? Start with chapter 1 of the Core Course. Already code? Jump into The Math Behind ML or Modern Architectures.

🧠 Core Course

4 chapters

  1. Chapter 1: What a Model Is: A Function with Knobs
  2. Chapter 2: One Neuron, Many Inputs: The Dot Product
  3. Chapter 3: Measuring Error: Choosing a Loss Function
  4. Chapter 4: Which Way Is Downhill? Derivatives by Hand and by Nudging

📐 The Math Behind ML

8 chapters

  1. Chapter 1: Linear Algebra, Vector Spaces & Matrix Calculus
  2. Chapter 2: Probability Distributions, Maximum Likelihood & MAP
  3. Chapter 3: Information Theory, Entropy & Divergence Measures
  4. Chapter 4: Convexity, Optimization Dynamics & Duality
  5. Chapter 5: Matrix Decompositions, SVD & Low-Rank Adaptation (LoRA)
  6. Chapter 6: Deep Neural Representations & Normalization Mathematics
  7. Chapter 7: Transformer Mathematics: Attention Scaling, RoPE & Latent Projections
  8. Chapter 8: Continuous Flow Matching ODEs & Post-Training RL Mathematics

💻 Modern Architectures

11 chapters

  1. Chapter 1: Linear Regression & Gradient Descent
  2. Chapter 2: Logistic Regression & Cross-Entropy Classification
  3. Chapter 3: Decision Trees & Information Gain (CART)
  4. Chapter 4: Multi-Layer Perceptrons & Analytical Backpropagation
  5. Chapter 5: Convolutional Networks & 2D Spatial Feature Maps
  6. Chapter 6: Modern Optimizers & Normalization Mechanics
  7. Chapter 7: Scaled Dot-Product & Multi-Head Attention
  8. Chapter 8: Modern Decoder-Only LLM Block (Llama / Mistral Style)
  9. Chapter 9: Mixture of Experts (MoE) & Multi-Head Latent Attention (MLA)
  10. Chapter 10: Diffusion Models & Rectified Flow Matching
  11. Chapter 11: Post-Training RL for Reasoning LLMs (DPO & DeepSeek-R1 GRPO)