Modern Artificial Intelligence Masterclass: Build 6 Projects

Build, train, and deploy two deep neural networks for emotion ai. A cnn with residual blocks predicts facial key points, while an emotion classifier detects emotions.

Explore emotion AI by building two neural networks to detect key facial points and classify expressions, then combine them for robust emotion inference from images.

Import libraries and datasets for facial keypoints detection using a cnn with residual blocks on 96 by 96 grayscale images, extracting 15 x,y points. Colab setup with TensorFlow 2.0.

Perform image visualizations of 96 by 96 facial images, overlay 31 x-y keypoints on random samples, and conduct a 64-image sanity check to validate data quality for AI training.

Perform image augmentation to expand the dataset by flipping images horizontally and vertically, adjusting brightness, and concatenating augmented data to improve model generalization.

Normalize and train on augmented 96x96 grayscale images, split data into training and testing (90/10), and prepare batch inputs for neural networks, with a mini challenge for 80/20.

Explore the theory and intuition of artificial neural networks, from neurons to the single neuron model. See how weights, bias, inputs, and activation functions—sigmoid, Relu, tanh—enable multi-layer perceptron learning.

Learn how artificial neural networks train with gradient descent to minimize loss, using training, validation, and testing splits to ensure generalization, and adaptive learning rate strategies.

Learn convolutional neural networks and residual networks, including feature extraction with kernels and pooling, then train dense classifiers and compare ResNets to AlexNet on ImageNet.

Build a deep residual network for facial key-point detection using Keras and TensorFlow 2.0, featuring conv and identity blocks, batch normalization, and max pooling insights.

Train a facial key points detector model with Adam optimizer and mean squared error loss, save the best model via checkpoint, and store weights in HDF5 and architecture in JSON.

Assess the trained ResNet facial points detection model on unseen data, load the best weights, and evaluate at about 85% accuracy, using GPU training for faster results.

Explore and preprocess a facial expression dataset for emotion classification, converting string pixel data to 96x96 images, and train a five-class model (anger, disgust, sad, happy, surprise) with visualization.

Visualize facial expression images and labels, perform exploratory data analysis, resize images to 96x96, and assess class balance with a seaborn bar chart to guide augmentation.

Prepare data and augment images for facial expression classification, splitting into train, test, and validation sets, normalizing pixels, and applying rotation, shifts, zoom, and flips.

Build and train a facial expression classifier using rest blocks and convolution blocks, with Keras data augmentation and two stages, for five emotions via softmax, with early stopping and checkpointing.

Explore how to assess classifier performance with a confusion matrix, accuracy, precision, and recall, distinguishing true and false positives and negatives, including type one and type two errors.

Assess a facial expression classifier by loading a pre-trained model from a json file and its weights, evaluating on test data, and examining accuracy, confusion matrix, and a classification report.

Combine facial key points and emotion models to produce joint predictions with about 85% accuracy, then visualize results for deployment.

Save a trained TensorFlow model in saved model format and deploy it with TensorFlow serving, enabling versioned servables, rest api access, and scalable inference for facial keypoints and emotions.

demonstrates serving trained models with tensorflow 2.0 serving by deploying two models, the key point model and the emotion model, on separate ports via tensorflow model server.

Explore how ai accelerates brain tumor detection and localization in mri scans by building a two-model pipeline: a tumor detector and a u-net segmentation model.

Apply deep learning to detect and localize brain tumors in MRI scans using a two-stage ResNet classifier and segmentation model, starting from problem understanding through evaluation.

Import libraries and datasets to build a brain tumor detection pipeline using ResNet for classification and a ResNet-based segmentation network for pixel-level tumor localization, confirming an unbalanced dataset.

Visualize brain MRI data and its binary tumor masks, using plotly and matplotlib, exploring unbalanced classes and side-by-side image and mask plots.

Explore the theory of convolutional neural networks and residual networks, including skip connections, and how CNNs extract features to achieve strong ImageNet performance.

Train a classifier model with transfer learning on a resnet backbone to detect brain tumors in MRI scans, then localize them with a resnet segmentation model, using data generators.

Load a pre-trained two-output classifier from json architecture and weights, use argmax on test predictions, compare to ground truth, and report 98% accuracy with confusion matrix and classification report.

Explore resnet-based segmentation by combining a u-net backbone with residual blocks to enable pixel-level brain tumor localization, using encoder, bottleneck, and decoder with skip connections.

Build a segmentation model to localize brain tumors using a ResNet-based encoder-decoder in Keras, training on MRI images with tumor masks, using skip connections and upsampling.

Train a segmentation resnet model to localize tumors using a focal tversky loss and a custom data generator, then save the best weights and architecture.

Load the pre-trained ResNet MRI architecture and segmentation weights, compile with Adam, and run the prediction pipeline to generate and compare predicted masks against ground truth for brain tumor localization.

Kick off a case study applying AI and ML to market segmentation, building targeted campaigns, and exploring exploratory data analysis with autoencoders, k-means, PCA, and Plotly visuals.

Explore AI applications in marketing through unsupervised learning and market segmentation. Build and evaluate clustering models with K-means, the elbow method, PCA, and autoencoders to tailor targeted campaigns.

Import core libraries (NumPy, pandas, Seaborn, Matplotlib, TensorFlow 2.0), load the sales dataset, and apply autoencoders with PCA and K-means for unsupervised market segmentation.

Learn exploratory data analysis and data cleaning in Python. Use Plotly Express for interactive bar plots and create dummy variables with get_dummies and one-hot encoding.

Perform exploratory data analysis and data cleaning on sales data using Plotly for interactive visualizations. Identify peaks, correlations, and trends with line plots, heatmaps, and scatter matrices.

Understand the theory and intuition of k-means clustering, an unsupervised algorithm that groups data by euclidean distance to centroids, using elbow method to choose clusters and within-cluster sum of squares.

Apply the elbow method to identify the optimal number of clusters for k-means by scaling data, plotting inertia across k values, and selecting the elbow point.

Apply the k-means clustering algorithm with five clusters to scaled sales data, obtain cluster centers and labels, inverse transform centers to original units, and visualize per-cluster distributions.

Apply principal component analysis to reduce features and visualize clustering in 2D and 3D, using k-means and PCA components to reveal distinct groupings and insights.

Explore the theory and intuition of auto encoders, including the encoder, code bottleneck, and decoder, for representation learning and dimensionality reduction.

Explore autoencoders for dimensionality reduction, then apply K-means elbow method and PCA to cluster data and reveal three meaningful customer segments.

Explore how ai and ml transform finance by predicting credit card defaults using AWS SageMaker Studio AutoML and XGBoost, building, training, and deploying models with hyperparameter optimization.

Frame a credit card default prediction as a binary classification problem using customer data, and explore building and tuning models with AWS SageMaker Studio, AutoML, and XGBoost.

Import libraries and the UCI credit card dataset to build a classifier predicting defaults. Describe the 30,000-sample data and features like limit balance, age, and encoded pay statuses.

Visualize and explore a credit card dataset using pandas, numpy, seaborn, and matplotlib; assess data quality, compare defaulted versus non-defaulted customers, and analyze distributions, correlations, and visualizations.

Create a cleaned training and testing dataset by one-hot encoding categorical features and combining them with numerical data for an XGBoost model.

Understand the theory and intuition behind XGBoost, an ensemble of gradient boosted trees learned from residuals, with many hyperparameters for fast, robust regression and classification.

learn the key steps of xgboost: build an initial model, compute residuals, train new trees on residuals, apply learning rate, and ensemble for scalable, generalizable gradient boosting.

Train and evaluate an xgboost classifier locally with sklearn, then deploy on AWS Sagemaker for inference, using train-test split and a grid search to tune hyperparameters.

Optimize xgboost hyperparameters with grid search in sklearn, testing gamma, subsample, colsample by tree, and max depth to select model and print classification report; explore aws sagemaker auto ml.

Explore how XGBoost works in AWS SageMaker and its gradient boosting ensemble. Tune hyperparameters such as max depth, eta, gamma, alpha, and lambda to optimize performance on tabular data.

Train an XGBoost model in AWS SageMaker by formatting data with a first target column. Save train and validation CSV files, upload to S3, and configure hyperparameters for the container.

Deploy a trained xgboost classifier on aws sagemaker, configure one m4.xlarge instance, and perform inference with text csv data. Evaluate precision, recall, and accuracy, and delete endpoint to avoid charges.

Train and deploy a binary classification model with AWS Autopilot, uploading data to S3; Autopilot performs feature engineering and data cleaning with minimal coding.

Explore creative ai to craft art masterpieces with the deep dream algorithm. Learn its theory and train a deep dream model using keras and tensorflow 2.0 to create trippy visuals.

Explore how creative AI can generate art, music, and stories, assess AI creativity against human benchmarks, and examine deep dream and neural networks through real-world art projects.

Import a pre-trained inception v3 model with top off and then on, explore activations and transfer learning, compare 21.8 million vs 23.8 million parameters, using TensorFlow 2.2.0 and Keras.

Blend two images, mars.jpg and eiffel.jpg, and preprocess the merged image for deepdream iterations to generate art. Normalize pixel values, expand and squeeze dimensions for batch processing, and visualize results.

Run a pre-trained Inception net v3, feed an image, extract activations from mixed three, five, seven, and beyond to build a deep dream feature extraction model and visualize layer outputs.

discover how the deep dream algorithm manipulates a pre-trained convolutional neural network to maximize layer activations by gradient ascent, revealing how early edges and deeper features shape dreamlike images.

Explore gradient operations in tf 2.0 with gradient tape to compute first-order derivatives, using y equals x cubed and y equals x to the fourth plus x to the fifth.

Implement the deep dream algorithm by calculating loss from layer activations and maximizing it through gradient ascent, focusing on mixed three, five, and seven activations.

Implement the deepdream step two by maximizing activations through gradient ascent on the input image using gradient tape and tf.function, with step size and steps, visualizing loss progress.

Apply the deep dream algorithm to generate a series of images from image zero, then assemble them into a video in task ten.

Build a food recognition classifier with AI using data robot to upload images, train models, and deploy, while exploring explainable AI for dessert, seafood, fried food, and vegetable and fruit.

Upload data to DataRobot, explore the dataset, and train an AI classifier that distinguishes four categories—fried food, seafood, vegetables and fruits, and dessert—and deploy the best model.

Train six ai models in datarobot, including a residual neural network with keras and tensorflow and a logistic regression with l1 and l2 regularization. Evaluate with confusion matrices.

Explore explainable AI with Grad-CAM visualizations in DataRobot, analyzing activation maps and heat maps from CNNs like Mobilenet V3 and ResNet, and preview deployment and predictions on new data.

Explore Amazon Web Services and cloud computing basics, including pay-as-you-go compute and storage with AWS S3 and SageMaker for training, deploying, and scaling machine learning models.

Explore the three key building blocks of AI systems—data, model, and compute—using AWS services like S3, SageMaker, and EC2 to store data, train models, and scale compute.

Understand AWS regions and availability zones, where regions host multiple availability zones as separate data centers with independent power and networking to ensure low latency, data compliance, and high durability.

Explore Amazon S3, a durable storage service for buckets and objects with fine-grained access controls. See storage classes, lifecycle policies, and a hands-on demo using SageMaker to train models.

Explore AWS EC2 and IAM essentials, learn how to select instance types for AI and ML workloads, leverage Sagemaker and elastic inference, and understand on-demand, spot, and reserved pricing.

Discover how to create a free AWS account and use the 12-month free tier to run EC2, store data in S3, and build, train, and deploy ML models with SageMaker.

Discover how AWS SageMaker streamlines the end-to-end machine learning workflow, from data labeling and preparation to model training, tuning, deployment, and scalable monitoring.

Discover how AWS SageMaker orchestrates the machine learning workflow—from data labeling with Ground Truth to notebooks, training, hyperparameter tuning, endpoints, and marketplace models.

Explore AWS SageMaker Studio as an integrated ML IDE that unifies model building, training, tuning, deploying, and managing experiments with Autopilot, debugging, and edge deployment.

Explore Amazon SageMaker Studio with a hands-on walkthrough of notebooks, kernel selection, and experiment workflows. Learn to upload data, run experiments, compare models, and deploy a production endpoint.

Explore aws sagemaker model training and deployment, covering training data in s3, training code in ec2 container registry, creating endpoints, and batch transforms for scalable inference.