Explore why artificial intelligence and machine learning matter, from optimizing large search spaces with metaheuristics to powering robotics, games, and self-driving technology through neural networks, deep learning, and reinforcement learning.

Explore supervised learning with labeled data, unsupervised learning with clustering, and reinforcement learning through environment interactions and rewards.

We explain feed-forward neural networks, their input, hidden, and output layers, and how neurons connected by weights learn through training.

Explore the fundamental feedforward neural network architecture with layered neurons and trainable weights, using backpropagation to achieve tasks like optical character recognition and object detection.

Activation functions introduce non-linearity. They enable non-linear boundaries and universal approximation, with sigmoid mapping to 0–1 for binary classification and tanh giving -1 to 1 and derivatives support training.

Learn how a feedforward neural network learns xor on a two-input data set, using an input layer, a hidden layer, sigmoid activations, and gradient descent with backpropagation.

Learn to minimize the neural network loss using gradient descent, updating weights via negative gradients with learning rate and momentum, guided by backpropagation for partial derivatives.

Learn how gradient descent and backpropagation use node delta and activation derivatives to update weights and biases across the forward and backward passes through hidden and output layers.

Explore the perceptron model with no hidden layers, using an input-output layer, activation outputs 0 or 1, and train with an error-based weight update and learning rate for xor.

Build a two-input perceptron network with an input and output layer, using a sign activation that outputs 0 or plus one, and initialize weights in [-0.5, 0.5] for 00,01,10,11 inputs.

Implement and test a perceptron network in Java, training with a simple and-like dataset, adjusting the activation threshold and learning rate until the error vanishes.

Explore why a single-layer perceptron cannot solve the xor problem, showing the need for a hidden layer and setting up for backpropagation in a feedforward neural network.

Explore how the perceptron model solves linearly separable problems while hidden layers enable deep neural networks to model nonlinear relationships.

implement a feedforward neural network with a hidden layer and sigmoid activation. organize input, hidden, and output layers with bias units, and train via forward pass and backpropagation.

Implement backpropagation in a Java neural network. Define activation function, constants, and layers; use sigmoid activation, learning rate 0.3, momentum 0.6, and 100000 iterations, deriving gradient from the sigmoid derivative.

Implement backpropagation in a neural network by building input, hidden, and output layers with edge weights, running forward passes, computing errors, and applying backpropagation with learning rate and momentum.

Implement the layer class for backpropagation, with private one dimensional arrays for inputs, outputs, weights, and weight changes, plus a bias node; initialize weights randomly in -2 to 2.

Implements the forward pass of a neural network layer, copying input to avoid mutating training data, computes weighted sums with bias, applies sigmoid activation, and returns a copied output.

Implement backpropagation to train a neural network in Java, calculating output and hidden layer deltas, updating weights with learning rate and momentum, and applying sigmoid derivatives.

Train a neural network to learn basic logical relations from input pairs (00, 01, 10, 11) and outputs for and, xor-like, and or patterns, highlighting how outputs differ by input.

Train a neural network with back propagation across many iterations, starting from random weights, and observe predictions improve toward correct OR results with more iterations.

XOR cannot be solved by a single layer perceptron, since it is non-linear; a feedforward backpropagation network with a hidden layer learns it, improving outputs across iterations.

Explore clustering with neural networks by using two input neurons for x and y coordinates to form color-coded clusters and apply K means clustering, adjusting training data.

Learn a concrete implementation of clustering with a feedforward neural network. Use two inputs, five hidden neurons, and three one-hot outputs for yellow, green, and blue clusters, trained by backpropagation.

Train a neural network on the iris dataset using four input features and backpropagation to classify iris setosa, versicolor, and virginica with a hidden layer.

Construct a neural network by transforming training data into a 2d array with one-hot outputs for three classes, using a Python script and backpropagation to train for optical character recognition.

Train a neural network on the iris data set using one-hot encoding for three classes, observe random weight initialization, and refine predictions through multiple iterations to achieve low error rate.

Measure neural network accuracy on the iris data set by splitting data into 70/30 training and test sets, training with backpropagation, and evaluating predictions against one-hot labels.

Implement optical character recognition using 5×5 pixel images into 25 inputs and 10 outputs, with a single neural network classifying digits zero through nine and 10-bit output encoding.

Download and install paint.net to manipulate pixels for optical character recognition, compare it with Windows built-in paint, and use Darknet to draw digits 0–9 from the chapter source.

Transform an 8x8 bitmap into a one-dimensional float array by mapping white pixels to 0 and black to 1, using a character reader to read the image.

Process 8x8 images into a 64-bit vector, encode digits 0–9 with outputs, and configure a neural network with 64 inputs, 15 hidden, and 10 outputs for optical character recognition.

Train a neural network for optical character recognition by transforming images to numerical data, using sigmoid outputs, and configuring 64 input, 15 hidden, and 10 output neurons.