# zekelabs/AI101-DeepLearning

AI101 - Comprehensive Deep Learning Tutorial
Switch branches/tags
Nothing to show
Latest commit c7fb4ce Dec 14, 2018

# Course Content

### Essential Programming

• Introduction to Deep Learning
• Introduction to Numpy
• Introduction to Tensorflow and Keras

### Essential basics of Linear Algebra

• Solution of Equations, row and column Interpretation

• Vector Space Properties

• Partial Derivative of Polynomial and Two conditions for Local Minima

• Physical Interpretation of gradient (Direction of Maximum Change)

• Matrix Vector Multiplication

• EVD and interpretation of Eighen Vectors

• Linear Independence and Rank of Matrix

• Orthonormal Matrices, Projection Matrices, Vandemonde Matrix, Markov Matrix, Symmetric, Block Diagonal

### Selected topics of Machine Learning

• Intuition behind Linear Regression, classification

• Grid Search

• Training Pipeline

• Metrics - ROC Curve, Precision Recall Curve

• Calculating Entropy

### Basics of Neural Network

• Evolution of Perceptrons, Hebbs Principle, Cat Experiment

• Single layer NN

• Tensorflow Code

• Multilayer NN

• Back propagation, Dynamic Programming

• Mathematical Take on NN

• Function Approximator

• Link with Linear Regression

• Dropout and Activation

• Optimizers and Loss Functions

### Introduction to Convolutional Neural Network

• 1D and 2D Convolution
• Why CNN for Images and speech?
• Convolution Layer
• Coding Convolution Layer
• Learning Sharpening using single convolution Layer in Tensor-Flow

### Different Layers in CNN pipeline

• Convolution
• Pooling
• Activation
• Dropout
• Batch Normalization
• Object Classification
• Creating Batch in Tensorflow and Normalize
• Training MNIST and CIFAR datasets
• Understanding a pre-trained Inception Architecture
• Input Augmentation Techniques for Images

### Transfer Learning

• Finetuning last layers of CNN Model
• Selecting appropriate Loss
• Adding a new class in the last Layer
• Making a model Fully Convolutional for Deployment
• Finetune Imagenet for Cats vs Dog Classification.

### Object Detection and Localization

• Different types of problem in Objects
• Difficulties in Object Detection and Localization
• Fast RCNN
• Faster RCNN
• YOLO v1-v3
• SSD
• MobileNet

### Autoencoders

• Image Compression - Simple Autoencoder
• Denoising Autoencoder
• Variational Autoencoder and Reparematrization Trick
• Robust Word Embedding using Variational Autoencoder

### Time Series Modelling

• Evolution of Recurrent Structures
• LSTM, RNN, GRU, Bi-RNN, Time-Dense
• Learning a Sine Wave using RNN in Tensorflow
• Creating Autocomplete for Harry Potter in Tensorflow

### GANs

• Generative vs Discrimative Models

• Theory of GAN

• Simple Distribution Generator in Tensorflow using MCMC (Markov Chain Monte Carlo)

• DCGAN,WGANs for Images

• InfoGANs, CycleGANs and Progressive GANs

• Creating a GAN for generating Manga Art

### Model Free Approaches in Reinforcement Learning

• Model Free Prediction
• Monte Carlo Prediction and TD Learning
• Model Free Control with REINFORCE and SARSA Learning
• Assignment Implementation of REINFORCE and SARSA Learning in Gridworld
• Off policy vs On Policy Learning
• Importance Sampling for Off Policy Learning
• Q Learning

### Behavioral Cloning and Deep Q Learning

• Understanding Deep Learning as Function Approximator
• Theory of Behavioral Cloning and Deep Q Learning
• Revisiting Point Collector Example in Unity and
• **Assignment : **Training Cartpole Example via Deep Q Learning

### Deep Learning in Action

• Face Detection using Yolo-v3
• Building Autocomplete Feature using RNNs
• Real-time Depth Prediction and Pose Estimation
• How is Deep Learning used in Autonomous Driver Assistant systems
• Tips and Tricks for scaling and easy Deployment of Deep Learning Models