misho-kr/Introduction to TensorFlow for Artificial Intelligence, Machine Learning, and Deep Learning.md

## Introduction to TensorFlow for Artificial Intelligence, Machine Learning, and Deep Learning.md

      
    Raw
  

              Introduction to TensorFlow for Artificial Intelligence, Machine Learning, and Deep Learning.md
            
          
    Introduction to TensorFlow for AI, ML, and DL

The "Machine Learning" course and "Deep Learning" Specialization from Andrew Ng teach the most important and foundational principles of Machine Learning and Deep Learning. This new deeplearning.ai TensorFlow Specialization teaches you how to use TensorFlow to implement those principles so that you can start building and applying scalable models to real-world problems.
Taught by

Laurence Moroney, AI Advocate, Google Brain
Week 1: A New Programming Paradigm

In week 1 you'll get a soft introduction to what Machine Learning and Deep Learning are, and how they offer you a new programming paradigm, giving you a new set of tools to open previously unexplored scenarios ... To get started, check out the first video, a conversation between Andrew and Laurence that sets the theme for what you'll study...


A Primer in Machine Learning
Rules and data go in answers come out. Rules are expressed in a  programming language and data can come from a variety of sources from local variables all the way up to databases.
Machine learning rearranges this diagram where we put answers in data in and then we get rules out.


The Hello-World of neural networks
Y = 2 * X + 1 (github) (colab)


Week 2: Introduction to Computer Vision

This week you’re going to take that to the next level by beginning to solve problems of computer vision with just a few lines of code!

Fashion-MNIST data

Train data and Test data
Data is a list of pairs of image and label


3-Layer neural network

Flatten: 28 x 28 -> 1D array
Dense: 128 neurons (think about is as a variable)
Dense: 10 outputs because tehre are 10 labels


model = tf.keras.models.Sequential([
  tf.keras.layers.Flatten(input_shape=(28, 28)),
  tf.keras.layers.Dense(512, activation=tf.nn.relu),
  tf.keras.layers.Dense(10, activation=tf.nn.softmax)
])

Workbook (colab)

Import tensorflow module
Normalize the data
Build the model
Compile the model
Tweak the NN
Use callbacks


Exercises

The output of the NN is a list of probabilities, the length of the list matches the number of labels
Increasing the number of neurons in the hidden layer improves somewhat the accuracy, but training takes longer
The dimension of the first layer must match the dimension of the input data
The range of training labels must match the number of output neurons
Increasing the number of hidden layers improves the accuracy but make the training take longer time
More training epouch improve the accuracy but only to some extent, sometimes the error increases due to overfitting
Skipping the normalization reduces the accuracy dramatically


MNIST Hardwriting Recognition workbook


References

Machine Learning Fairness


Week 3: Enhancing Vision with Convolutional Neural Networks

In week 2 you saw a basic Neural Network for Computer Vision. It did the job nicely, but it was a little naive in its approach. This week we’ll see how to make it better, as discussed by Laurence and Andrew here.

Convolution is a way to condense the image down to the important features, for example Conv2D
Pooling is a way of compressing an image, for example MaxPooling2D

model = tf.keras.models.Sequential([
  tf.keras.layers.Conv2D(64, (3,3), activation='relu', input_shape=(28, 28, 1)),
  tf.keras.layers.MaxPooling2D(2, 2),
  tf.keras.layers.Conv2D(64, (3,3), activation='relu'),
  tf.keras.layers.MaxPooling2D(2,2),
  tf.keras.layers.Flatten(),
  tf.keras.layers.Dense(128, activation='relu'),
  tf.keras.layers.Dense(10, activation='softmax')
])


Improving Computer Vision Accuracy using Convolutions workbook (colab)


Walking through convolutions workbook (colab)


References

YouTube Videos on Convolutions
Lode's Computer Graphics Tutorial


Week 4: Using Real-world Images

Last week you saw how to improve the results from your deep neural network using convolutions. It was a good start, but the data you used was very basic. What happens when your images are larger, or if the features aren’t always in the same place? Andrew and Laurence discuss this to prepare you for what you’ll learn this week: handling complex images!

What happens when you use larger images, millions of colors and where the feature might be in different locations of the image
Image generator generates labels from the subdirectories that contain the images, it also resizes at runtime

from tensorflow.keras.preprocessing.image import ImageDataGenerator

# All images will be rescaled by 1./255
train_datagen = ImageDataGenerator(rescale=1/255)

# Flow training images in batches of 128 using train_datagen generator
train_generator = train_datagen.flow_from_directory(
        '/tmp/horse-or-human/',  # This is the source directory for training images
        target_size=(300, 300),  # All images will be resized to 150x150
        batch_size=128,
        # Since we use binary_crossentropy loss, we need binary labels
        class_mode='binary')


More convolution layers and more training epochs to deal with higher image sizes workbook (colab)


Sigmoid activation function is suitable for binary classification tasks


Compacting images, i.e. reducing image size and/or eliminating convolution layers -- workbook (colab)

Improves the training accuracy (likely due to overfitting)
Reduces the validation accuracy


References

Machine Learning - Stanford University | Coursera -- Learning Rate
Binary Crossentropy
RMSPropOptimizer
Overview of mini-batch gradient descent
Binary Classification
Pixabay