Skip to content

🧠 01 — Neural Network Intuition

A neural network is a stack of layers that learns transformations from inputs to outputs.


Basic Components

Component Meaning
Neuron computes weighted sum + activation
Layer group of neurons
Weight learned importance value
Bias learned offset
Activation nonlinear function
Loss error measure

Why Nonlinearity Matters

Without activation functions, stacked layers behave like one linear model.

Common activations:

  • ReLU
  • sigmoid
  • tanh
  • softmax

Use Cases

  • image classification
  • NLP
  • speech
  • recommendation systems
  • complex nonlinear prediction

For small tabular data, tree models often beat neural networks.


Next

➡️ 02-training-neural-networks