Artificial Intelligence Bootcamp in R Programming

Learn to read csv and excel data into R, set the working directory, and handle separators and headers using read.csv, read.table, and read_excel.

Learn to read data from online HTML tables, such as Wikipedia's 2016 Summer Olympics medal table, directly into R using readHTMLTable and URL retrieval.

Read data from online html tables using the read_html function, target tables with the X part, and load results into variables for Wikipedia Olympics and UK World Heritage pages.

Learn to read external data in the H2O framework by importing files, setting the working directory, and initializing an H2O server, then split data into 75/25 training and testing sets.

Learn to handle missing data in R by identifying NA values, removing rows with missing values (complete cases), and using mean substitution or advanced imputation with the mice package.

Learn how to clean and prepare data in R by removing dollar signs and commas from GDP per capita, converting to numeric, and selecting and renaming columns.

Discover how to use dplyr for data summarizing and exploration with air quality data; perform selection, filtering, mutation, and grouping to compute mean ozone and monthly wind averages.

Learn to use dplyr piping to select ozone and month from air quality, then summarize mean ozone and temperature by month with group by and filter wind greater than twelve.

Perform basic exploratory data analysis in R on tweet polarity data from the Jerusalem Embassy dataset, including text cleaning, a document matrix, frequent terms, and k means clustering.

Identify the main data types used in statistical analysis: categorical, numerical, and ordinal data. Explore surveys, measurements, and rating scales to see how each type informs analysis.

Explore the core theories of machine learning, contrasting supervised and unsupervised learning, with practical insights on classification, regression, training data, and remote sensing imagery analysis.

Explore neurons and activation functions, examine how neural networks work with a housing-price example, and master gradient descent, stochastic gradient descent, and backpropagation in an ai bootcamp in R programming.

Explore the neuron as the building block of artificial neural networks. It maps input values through weights and an activation function across input, hidden, and output layers to produce predictions.

Identify activation functions used in networks, including threshold, sigmoid, rectified activation function, and hyperbolic tangent. Understand how these functions enable hidden layers to produce probabilistic outputs in the final layer.

Explore how a trained neural network estimates property valuation by processing input features through input, hidden, and output layers with weights and activation functions.

Explore how neural networks learn by contrasting hardcoded rules with data-driven training, using a one-layer perceptron, activation functions, cost functions, and backpropagation to minimize error.

Learn gradient descent minimizes the neural network cost function by iteratively adjusting weights after backpropagation, avoiding brute-force search amid the curse of dimensionality.

Explore stochastic gradient descent and how it updates weights after every row, avoiding local minima and enabling faster training than batch gradient descent for neural networks.

Explore how backpropagation optimizes neural networks by adjusting all weights simultaneously through forward propagation, error calculation, and gradient-based updates, including stochastic and mini-batch gradient descent.

Explore binary classification with neural networks to predict credit card defaults using predictors such as payment history, bill amount, age, education, gender, and marital status.

Explore confusion matrices and key accuracy metrics for binary classification, including true positives, true negatives, false positives, and false negatives, plus precision, recall, and Cohen's kappa.

Implement a multi-layer perceptron for supervised classification in R, including data preparation, train-test split, normalization, backpropagation with logistic activation, and evaluating with predictions and confusion matrices.

Explore neural networks for multiclass classifications with three or more levels, using iris species and loan status data, apply softmax modeling, generate predictions, and evaluate accuracy with confusion matrices.

Explore a multi-class image classification using neural networks to recognize handwritten digits 0–9. Learn to load image data, prepare pixel predictors, and predict unseen labels.

Learn to perform multi-class classification on imagery type data using neural networks with the caret package, classifying 28 by 28 pixel fashion item images.

Implement an artificial neural network with the H2O architecture for multi-class clothes image classification, using 785 pixel predictors and one hidden layer, achieving about 77% accuracy.

Apply MXNet to build an artificial neural network for binary classification, creating and training a multi-layer neural network, and evaluating performance with soft max activation and accuracy metrics.

Implement a keras mlp to classify MNIST digits. Use a 784-input network with a 784-unit hidden layer, dropout 0.4, and a 10-unit softmax output, trained on data for 100 epochs.

Train a keras MLP on the iris data frame with a 70 percent training and 30 percent testing split, using one-hot encoding and normalized features for robust multi-class classification.

Demonstrates building a Keras sequential mlp for iris regression to predict petal length from sepal length, train 125 samples and test 25, using leaky ReLU, dropout, mean squared error.

train a neural network to predict income from education, gender, prestige, and profession, evaluate RMSE on a 75/25 split, and note education dominates while prestige shows little influence.

Explore neural network regression with the neuralnet package to predict cement strength from predictors such as cement, slag, ash, water, and aggregates. Test a one-hidden-neuron model on training data.

Implement regression with an ANN in MXNet by building a fully connected hidden layer and a linear regression output, trained on Boston housing data and evaluated by mean square error.

Explore identifying variable importance in neural networks using neural interpretation diagrams and Carson's and Olden algorithms, applying them to cement strength data with predictors like super plastic, water, and aggregates.

Train a neural network with two hidden layers using the neuralnet package to classify colleges as private or not, using normalization and a 70/30 split, achieving 93% accuracy.

Implement a simple deep neural network for regression in R with the deepnet package, featuring two hidden layers, a Boston data train-test split, scaling, and MSE evaluation.

Build a deep neural network with the H2O architecture for multi-class fashion item classification, using a 75/25 train-test split, 3 hidden layers, and dropout.

Implement a less intensive deep neural network in R with H2O for supervised classification, achieving about 88% accuracy with two hidden layers of 162 to 200 neurons.

Apply two hidden layers (162 neurons each) to achieve about 88 percent accuracy on a 75/25 train-test split using H2O, cross-entropy loss, and multi nominal classification.

Learn to build a deep neural network with two hidden layers using Keras to classify glass types from real data, with data preparation, one-hot encoding, normalization, and evaluation.

Identify the most important predictors for distinguishing malignant from benign tumors using deep learning with h2o, including data cleaning, variable importance plots, and high AUC performance.

Implement mxnet via the caret package in R to build a three-layer neural network for malignant or benign classification, train on a 75/25 split, and evaluate with a confusion matrix.

Implement a deep neural network for regression using the H2O deep learning algorithm on the Boston housing dataset, predicting house prices from 13 predictors.

This lecture demonstrates implementing a DNN for regression in Keras using the Boston housing data, with normalization and a two-layer 64-neuron network.

Implement deep neural network regression with Keras on a data set of Ames housing prices, using 70/30 splits, normalized features, and a sequential 64-unit model evaluated by mean absolute error.

Explore how unsupervised classifications cluster pixels by spectral properties in remote sensing data to reveal land-use patterns.

Explore unsupervised learning with autoencoders to uncover underlying data patterns using deep neural networks, demonstrated through credit card fraud detection and cancer detection.

Implement an autoencoder in R with H2O to detect credit card fraud, loading data, converting to H2O format, and training a three-hidden-layer unsupervised model on predictors 1–30.

Apply an autoencoder for anomaly detection on credit card fraud data using unsupervised learning with numerical predictors and reconstruction error to separate fraud from nonfraudulent cases.

Learn to use autoencoders for unsupervised classification by extracting deep features to separate binary and malignant tumors, using layer 3 representations and visualizing with a cube plot.

Learn to build autoencoders with Keras to compress 28x28 grayscale MNIST images into 32-neuron encodings and reconstruct 784-dimensional vectors in an unsupervised setup.

Implement autoencoders on real credit card transaction data to detect fraud without labels, using a 70/30 train-test split, normalization, and a dense symmetric encoder architecture in R with Keras.

Explore stacked autoencoders in Keras, building multi-layer encoders and decoders to model credit card data, achieve high accuracy, and distinguish fraudulent transactions.

Detect outliers using a Keras autoencoder on a 28x28 image dataset, preparing train and test sets, reshaping to 784-length vectors, scaling to 0-1, and evaluating loss during training.

Detect outliers with an autoencoder in R by reconstructing the test set, measuring reconstruction error, and using a threshold to classify normal values and outliers, achieving 78 percent accuracy.

Apply autoencoders to detect outliers in a cancer dataset using Keras in R, distinguishing malignant from benign tumors by reconstruction error and a tunable threshold.

Explore what volitional neural networks are, compare brain and artificial networks for image recognition, and outline a stepwise plan to build a convolutional neural network from feature detectors to softmax.

Explore how humans and machines recognize images by processing features, and learn how convolutional neural networks recognize visuals through convolution, pooling, and fully connected layers.

Explore step 1 of convolution in deep learning by applying feature detectors (filters) to an input image, producing feature maps with stride while preserving spatial relations.

Explore how the ReLU layer adds non-linearity on top of convolution to create feature maps by rectifying negative values to zero.

Explore max pooling and how pooling preserves features while achieving spatial invariance, reducing feature map size and lowering parameters with 2x2 windows and stride 2.

Flatten the pooled feature maps into a single column vector to feed an artificial neural network after convolution and activation of rectified linear units with pooling.

Add a fully connected network to convolutional pipeline, turning flattened features into dog and cat outputs. Train with backpropagation and gradient descent to optimize weights and feature detectors, reducing loss.

Explore how convolutional neural networks use feature detectors and max pooling to extract and condense features, then flatten and classify with a fully connected network trained by forward and backpropagation.

Explore how the softmax function converts neural network outputs into probabilities that sum to one, and how cross-entropy loss drives classification learning in neural networks.

Develop a CNN in R that uses two convolution layers, pooling, and fully connected layers to classify fashion images, achieving about 89 percent test accuracy in multi-class classification.

Explore a convolutional neural network for fashion item classification, visualize its layered architecture, and analyze training and testing accuracy to refine performance on unseen data.

Build a convolution neural network with Keras using a sequential model for 28x28 grayscale images. Normalize data, train on 60,000 images, and reach around 98% accuracy after five epochs.

Train a CNN with Keras and evaluate on test images and labels. Track accuracy and loss across epochs with 80/20 validation split, then save and load the model for reuse.

Build a keras cnn to classify chest x-ray images as normal or pneumonia using a real dataset, with augmentation, image generators, and train/validation/test folders.

Train a CNN on medical images with augmentation and rescaling to reduce overfitting, monitoring training and validation accuracy; use a multi-layer architecture with max pooling and directory-based data.

Improve cnn performance by tackling overfitting with data augmentation and dropout, monitor training and validation accuracy, and optimize epochs to boost testing accuracy.

Learn how to pre-process text data in R by turning an email column into a corpus, then lowercase, remove punctuation and stop words, apply stemming, and build a document-term matrix.

Detect fraud in text data using a Keras autoencoder trained on a preprocessed document-term matrix. Split data 70/30, set an outlier threshold to flag suspicious emails.

Explore word embeddings to classify fraud in Enron emails, using a deep learning model with an 8-dimensional embedding, 10,000 word vocabulary, and 100-word max length, achieving around 80% validation accuracy.

Harness GloVe word embeddings to classify emails in the Enron dataset using unsupervised learning-derived vector representations, converting text to sequences and building a Keras sequential model for supervised classification.