Neural Network

What Neural Network is

A neural network is loosely inspired by biological neurons but is really a stack of matrix multiplications with non-linearities in between. Weights are the learned parameters; the architecture decides how information flows between layers.

How it works

Inputs are encoded as numbers, multiplied by weight matrices, offset by biases and passed through an activation function. Training compares the output to the target and updates weights through backpropagation until the loss stops improving.

Why it matters

Nearly every modern AI capability is a neural network underneath. Knowing what a layer, weight and activation are makes the rest of the field readable.

Common uses

• →Classification and regression on tabular data
• →Image and audio recognition
• →Text embeddings
• →Control policies in robotics