Practical Deep Learning Contents

Day 1

Introduction

Introduction to Artificial Intelligence and the focus of the class

Linear Regression

Linear Regression definition, how to create models, how to train linear regression models, how to generate data, how to test accuracy of regression models

Exercise 1

Linear Regression using standard Python

Perceptron

Classification definition, basic neuron definition and operation, neuron creation, neuron training, accuracy testing

Exercise 2

Implementing a Perceptron using standard Python

Multi-Class Models

Classification using multiple neurons, multiple neuron models, multiple neuron error calculation, multiple neuron optimization, multiple neuron training

Exercise 3

Implementing a multi-class classification model using standard Python

Deep Neural Networks

Activation functions, multiple layer network creation, multiple layer operation, multiple layer optimization, multiple layer error propagation, loss differentiation

Exercise 4

Implementing a Deep Neural Network using standard Python

Day 2

Artificial Intelligence Frameworks

Introduction to AI frameworks, Introduction to Tensorflow

Tensorflow Linear Regression

Placeholders, Variables, Sessions, Running sessions, optimization, calculating error

Exercise 5

Implementing Linear Regression of housing prices using TensorFlow

Tensorflow Classification

Softmax definition, Cross Entropy definition, error propagation, Tensorflow Optimizers, Learning rate, epochs

Exercise 6

Implementing MNIST classification using TensorFlow

Tensorflow Deep Networks

Tensorflow activation functions, Tensorflow differentiation, Tensorflow error propagation, Tensorflow Deep Network optimization, Deep Network descriptions

Exercise 7

Implementing MNIST classification using a Deep Neural Network with TensorFlow

Sessions, Graphs, Saving and Restoring

Session and graph description. Running sessions on multiple platforms including CPU, GPU, Mobile/Embedded. Save a graph, restore a graph, using a graph for inference only.

Exercise 9 and 9a

Training an MNIST classifier network, save a checkpoint, restore the checkpoint in a new session and evaluate results

Day 3

Visualizing Model Operation

Tensorboard, adding summaries, adding histograms, adding graphs, interpreting results

Exercise 10

Using Tensorboard to visualize training and results

Convolutional Neural Networks

Convolutional filters, feature maps, convolutional layers, pooling layers, fully connected layers, stride, padding, constructing CNN networks, training CNN networks

Exercise 11

Implementing a CNN network using tf.layers 

Transfer Learning

What makes a good data set, balanced data sets, distinct data sets, non-conflicting data, ImageNet, Inception-V3, transfer learning description, transfer learning operation, transfer learning

Exercise 12

Retrain Inception-V3 with new training data

Object Detection

Detecting multiple objects in an image, bounding boxes, different types of object detection algorithms, object detection transfer learning

Embedded Vision

TensorFlow Lite, weight quantization, operation on mobile/embedded devices