Skip to content
AI.info

AI.info

Neural networks

Differentiable programs, traced by hand from a single neuron to an architecture you can defend in a review.

  1. Neural Networks as Differentiable Programs
  2. Inside an Artificial Neuron
  3. Perceptrons and the Geometry of Linear Separation
  4. Dense Layers: Many Neurons, One Matrix
  5. Tensors, Axes, Shapes, and Batches
  6. Activation Functions: Where Nonlinearity Enters
  7. Depth, Composition, and Representation Hierarchies
  8. Parameter Sharing and Inductive Bias
  9. The Forward Pass and Computational Graphs
  10. Output Heads: Connecting a Network to a Task
  11. Loss Functions as Learning Contracts
  12. Gradients: Local Sensitivity in Parameter Space
  13. The Chain Rule and Credit Assignment
  14. Backpropagation, Worked From Output to Input
  15. Automatic Differentiation in Practice
  16. The Training Loop: From Mini-Batch to Parameter Update
  17. Initialization and the Scale of Signals
  18. Vanishing, Exploding, Saturated, and Dead Gradients
  19. Normalization and Train–Evaluation Behavior
  20. Residual Connections and Stable Depth
  21. Capacity, Regularization, Dropout, and Weight Decay
  22. Embeddings and Learned Representation Spaces
  23. Convolutions: Locality, Reuse, and Equivariance
  24. Recurrence: State, Memory, and Sequence Order
  25. Attention: Content-Based Interaction
  26. Architecture Motifs: Encoders, Decoders, Bottlenecks, and Towers
  27. Reading Hidden Representations Without Overclaiming
  28. Debugging and Testing Neural Networks
  29. Capstone: Design, Trace, and Defend a Neural Network