Neural Networks Introduction

Neural Network Basics

Neural networks are flexible models inspired by brain neurons.

Perceptron

Single neuron: inputs × weights + bias. Activation function: step, sigmoid, ReLU.

Multiple outputs for multi-class classification.

Multi-Layer Perceptron

Hidden layers between input and output. Each layer transforms representations.

Backpropagation computes gradients. Gradient descent updates weights.

Architecture

Number of layers and neurons matters. Too few underfits; too many overfits. Regularization helps.

Key Takeaways

1. Neural networks learn representations
2. Backpropagation enables training
3. Architecture affects capacity