5 Neural Networks Free Online FlashCards

What is a neural network?

A machine-learning model made of connected computational units that transform inputs into predictions.

How does an artificial neuron transform its inputs?

It computes a weighted sum plus a bias, then applies an activation function: $z = \sum_{i=1}^{n} w_i x_i + b$, $a = \phi(z)$.

What roles do weights and biases play in a neuron?

Weights determine how strongly inputs contribute, while biases are added to the weighted sum. Both are learned parameters.

What are the roles of a network’s three layer types?

The input layer receives features, hidden layers transform information, and the output layer produces the prediction.

What happens during a forward pass?

A forward pass sends values through the layers in order until the network produces an output.

Why do layered networks need nonlinear activation functions?

Without nonlinear activation functions, composing linear transformations remains linear, so adding layers alone cannot represent nonlinear patterns.

What information do supervised training examples contain?

In supervised learning, training examples pair inputs with their correct target values.

What does a loss function measure during training?

A loss function compares predictions with target values to measure prediction error.

What does backpropagation calculate?

Backpropagation works backward through the network to calculate how parameter changes affect the loss, producing gradients.

How does an optimizer use gradients during training?

An optimizer uses gradients to update weights and biases in a direction that reduces loss; gradient descent is a common method.

What is an epoch?

An epoch is one full pass through the training examples.

What makes a neural network deep?

A deep neural network has multiple hidden layers; deep learning is the training of such networks.