TensorFlow - An open source Machine
Learning Course

Introduction

• Introduction to TensorFlow
• Architecture of Tensorflow
• Installation on Local Machine
• Using TensorFlow in Google Colab
• Working with Tensors and Operations
• Keras Low level api
• Labs:-
• 1. Installing tensorflow on local machine 2. Working with google colab

Introduction to Artificial Neural Networks

• a. From Biological to Artificial Neurons
• b. Different activation functions
• c. What is perceptron?
• d. Multilayer perceptron and back propagation
• e. Working with sequential api
• f. Working with the functional api
• g. Using callbacks
• h. Tensorboard for visualization
• i. Labs:-
• 1. Classification on iris dataset using perceptron model
• 2. Image classification on fashion MNIST dataset using sequential and functional API
• 3. Improving the model using callbacks

Training Deep Neural Nets

• a. Challenges of Deep Neural Networks’
• b. Vanishing and Exploding gradients
• c. Glorot and He Initialization
• d. Non Saturating Activation Functions
• e. Different activation functions effect on deep neural nets
• f. Batch normalization
• g. Reusing the pre-trained layers in Neural nets
• h. Faster optimizers
• i. L1 and L2 regularization
• j. Dropouts and their purposes
• k. Labs:- 1. Classification of MNIST dataset and performance of models by the use of
• different activation functions.
• 2. Classification by using a pre-trained layer of model to reduce the computation time.
• 3. Reducing the overfitting through the use of regularization in various models.

Loading and preprocessing the data

• a. The data api
• b. Chaining transformations
• c. Pre-processing the data
• d. TFR record format and compressed files
• e. Introduction to protocol buffer
• f. Processing the Input features
• g. TF transform
• h. Tensorflow datasets project
• i. Labs:- 1. Implementation of basic functions like repeat, batch, shuffle required for
• preprocessing.
• 2. Implement interleave() so as to read many files at a time using tensorflow for improving
• the performance.
• 3. Storing and accessing the files stored in tfrecord format for better processing.

Computer Vision using CNN

• a. Inspiration to the CNN
• b. Architecture of CNN
• c. Convolution layers in CNN
• d. Filters in CNN
• e. Pooling layer in CNN
• f. Depth pooling in CNN
• g. Different architectures of CNN
• h. Labs:- 1. Implementing different filters to find the different patterns from the image.
• 2. Progressively reduce the spatial size of an image.
• 3. Classifying the fashion MNIST dataset by using CNN.

Processing Sequences using RNN

• a. Single neuron RNN
• b. Working with RNN neural network
• c. Input and Output Sequences in RNN
• d. Introduction to Deep RNN
• e. Forecasting using RNN
• f. Unstable gradient problem
• g. Architecture of LSTM
• h. Architecture of GRU
• i. Labs:- 1. Forecasting using simple RNN, Deep RNN, LSTM and GRU

Natural Language Processing

• a. Introduction to Natural Language processing
• b. Shakespear text generation using char RNN
• c. Stateless and stateful RNN
• d. Concept of sentiment analysis
• e. Encoder and Decoder Network for Neural Machine Translation
• f. Pre-processing required for encoder and decoder
• g. Concept of Beam Search
• h. Overview of attention mechanism
• i. Labs:- 1. Generating Shakespearean text using character RNN, Bidirectional RNN.
• 2. Sentiment analysis of IMDB dataset.

Representative Learning and Generative Learning using Autoencoders and GANs

• a. Introduction to Autoencoders and GAN
• b. Efficient data representation
• c. Dimensionality reduction using autoencoders
• d. Introduction to stacked autoencoders
• e. Training one autoencoder at a time
• f. Recurrent autoencoders
• g. Sparse autoencoders
• h. Generative adversarial networks
• i. Deep Convolutional GANs
• j. Labs:- 1. Autoencoder for the dimensionality reduction
• 2. Reconstruction of images using recurrent auto encoder

Reinforcemnent Learning

• a. Introduction to reinforcement learning
• b. Learning to optimize rewards
• c. Policy Search
• d. Introduction to OpenAI Gym
• e. Neural Network Policies
• f. Credit Assignment Problem
• g. Markov Decision Process
• h. Q Learning
• i. Deep Q Learning
• j. Labs:- 1. Train a Deep Q network with TF agent(Cartpole Environment)