Course Outline

Module 1: Introduction to Theano

• What is Theano?
• Theano code – Basics
• Graph Visualization
• Variables, constants, functions, and Placeholders
• Theano Basic operations in Neural Networks
• Case Study: Building a simple neural network using Theano to predict house prices.

Module 2: Introduction to Neural Networks

• Understanding Neural networks
• Supervised and unsupervised learning with Neural Networks
• Logistic and linear regression with Theano
• Understanding vectorization in Neural Networks
• Working with Activation functions
• Model Training and Validation
• Forward and Backward propagation
• Understanding the perceptron concept
• Case Study: Using a neural network to classify handwritten digits from the MNIST dataset.

Module 3: Improving Neural Networks

• Understand the optimization algorithm
• Regularizing your Neural Network
• Learn the tuning process
• Work with Hyperparameter Tuning
• Understand Normalization in Neural Networks
• Fitting and testing the Batch Norm into a Neural Network
• Understand the Softmax Regression
• Case Study: Fine-tuning a neural network to improve accuracy for customer churn prediction.

Module 4: Convolutional Neural Network for Data Visualization and Analysis

• What is CNN?
• Understanding pooling layers
• Edge Detection
• Building CNN components
• Build a full CNN and test on SVHN
• Analyze a model’s training performance in Theano
• Identify cases of overfitting and apply techniques to prevent it
• Visualizing learned filters
• Working with a XOR problem
• Perceptron Networking
• Case Study: Applying a CNN to classify traffic signs for autonomous driving systems.

Module 5: Feed Forward Network for Data Visualization and Analysis

• Introduction to Feed Forward Network
• Combination of units to layers and networks
• Understanding Non-linear Neural Networks
• Understanding various activation functions used
• Working with a XOR problem
• Gradient-based learning
• Input-output mapping
• The Applications of the Feed Forward Network
• Case Study: Predicting stock prices using a feedforward neural network.