Artificial Intelligence-Computational Intelligence in Python

Examine fuzzy logic systems that map crisp inputs to fuzzy sets with membership degrees using a rule base and inference engine. Learn how defuzzification converts aggregated outputs into crisp results.

Explore a fuzzy inference system that models tipping based on service and food quality, using triangular membership functions, centroid method to obtain a crisp tip value, and Python implementation.

Explore fuzzy rules that map input variables food and service to output levels low, medium, and high through fuzzification, membership functions, and rule implication.

Aggregate the output membership functions with the max operator across low, medium, and high activations, then perform defuzzification using the centroid method to obtain the crisp output.

Explore how a fuzzy inference system uses the centroid method to produce interpretable outputs, such as a 20.99% tip, and learn about optimization with genetic algorithms.

Explore how evolutionary algorithms solve optimization problems by evolving a population of candidate solutions, using fitness, selection, crossover, and mutation to improve over generations.

Explore how a genetic algorithm initializes a population, encodes genes into chromosomes, and uses fitness evaluation, selection, crossover, mutation, and elitism to evolve optimal solutions.

Implement a genetic algorithm to optimize a six-input function using a fitness function that compares the produced output to the desired output, employing the Bigart library for population evaluation.

Initialize a population for a genetic algorithm by setting the number of chromosomes and six genes per chromosome, choosing a -2 to 5 range, and filling with random uniform values.

Define a genetic algorithm constructor with generations, population, fitness function, genes, and range, using roulette selection, single-point crossover, and mutation capped at 10%; run, identify top solution, and predict output.

A genetic algorithm optimizes a solution vector to predict outcomes, with best values around 0.3309 and 0.776, a plateau near 60-65 generations, and an output near forty four.

Develop a neural network to recognize digit labels from binary input patterns, using 20 inputs and 20 hidden nodes, scaling outputs by dividing by 10 to map 0–9.

Define the neural network class in Python with input, hidden (size 20), and output layers; initialize weight matrices with a Gaussian mean zero and apply sigmoid activation and its derivative.

Define a make predictions procedure to reshape input to five by four, perform a forward pass, and train the network for 5,000 epochs, reporting mean squared error.

Observe how the neural network's loss decreases through epochs using backpropagation, forward and backward passes, and training patterns for prediction, paving the way for hybrid intelligent systems.

Explore the Mackey-Glass chaotic time series governed by a differential equation, and learn how the previous four time stamps predict future values, specifically X plus six.

Download the dataset directory, install the required package, import the membership function libraries, read the time-series training data, and construct input and output variables for subsequent membership function design.

This lecture examines how a two-input system computes errors with brackets and membership functions, compares real versus predicted outputs, and notes performance limits with sharp kinks.

Four-input configurations with added membership functions for each input improve performance, adjust the filename and variables, and reduce the error to 0.0324.