Deep Learning: Visual Exploration

Deep neural networks visually explained in plain english & without complex math

Created byVladimir Grankin

Last updated 4/2018

English

What you'll learn

Deep understanding of what is deep neural network and how exactly it works under the hood to come up with good predictions in real life problems (we will only explore feedforward deep neural network for binary classification in our course, but we discuss fundamentals so knowledge you will get is also applicable to all the other network types!)
Understand what is decision boundary and how exactly it is formed by a deep neural network
Understand why deep neural networks are also knows as function approximators

Course content

1 section • 15 lectures • 4h 1m total length

Before We Begin6:54
In this video I will provide you some links to good materials you may need in order to get up and running with deep learning.
What We Will Do1:45
High altitude overview of what we will do in this course.
Visual Exploration - Tensorflow Playground15:46
Here I will guide you through famous tensorflow playground platform which allows us to create simple deep neural networks and visually see how they work.
Preparing Data11:42
In order to train our neural network we first need data set to work with. In this video we will see data we will work with.
Creating Basic Deep Neural Network (DNN)21:39
In this video we will create a simple so called "feedforward" deep neural network for binary classification. This will be a network we will use throughout the course.
Visual Exploration - Sigmoid Function (Activation Function)38:02
In this video we will see what is activation function and why it is used by a neural network. In particular we will explore sigmoid function, every single aspect of it.
Finding Parameters (Training our Deep Neural Network)11:33
In this video we will train our neural network using simple trick instead of backpropagation in order to avoid complex math and make our experiments easier for the beginner.
Visual Exploration - Signal Journey Through Deep Neural Network (2D)27:03
In this video we will see how input data is travelling through neural network, from start to end, in 2D.
Backpropagation Intro15:00
In this video I will briefly introduce you to backpropagation. We will discuss what it is and what it does.
Visual Exploration - Signal Journey Through Deep Neural Network (3D)16:13
In this video we will continue our journey on how input signal is transformed as it goes from neural network's input to it's output. This time in 3D.
Visual Exploration - Visualizing Neurons of Our Deep Neural Network30:49
In previous videos we visualized data transformation. But data is transformed according to functions used in neurons. So in this video we will visualize functions themselves!
Visual Exploration - Slices 2D (Neural Network's Magic Behind Good Predictions)15:54
In this video we will see what is the magic behind deep neural network. We will visually see why and how neural network is able to make accurate predictions using. We will use 2D plots.
Visual Exploration - Slices 3D (How Exactly Neural Network Makes Predictions)13:32
In this video we will continue to explore the magic behind deep neural network. In 3D this time.
Visual Exploration - Decision Boundary (How Deep Neural Network Classifies Data)6:50
In this video you will see how to visualize decision boundary for your neural network (just like the decision boundary you see in tensorflow playground neural network constructor).
Summary & Homework8:51
Summary and ideas for you to try at home.

Requirements

Basic python skills - optional, you will need python if you want to run the code we discuss yourself
Familiarity with neural networks at high level (terms like bias, weight and activation function should be familiar)
Jupyter notebook (optional and is needed if you want to run all the demos yourself)

Description

Visual introduction to Deep Learning based on simple deep neural network. Take this course if you want to understand the magic behind deep neural networks and to get a excellent visual intuition on what is happening under the hood when data is travelling through the network and ends up as a prediction at it's output.

In this course we will fully demystify such concepts as weights, biases and activation functions. You will visually see what exactly they are doing and how neural network uses these components to come up with accurate predictions.

Who this course is for:

People who like to understand things visually
People who like simple explanations against mathematical and formal ones
If you are just starting with Deep Learning or AI in general, this course if for you!
If you think what is happening under the hood of deep neural network is a mystery, this course is for you! - we totally demystify DNNs in this course!
If you wonder how exactly weights, biases and activation functions are working, this course is for you!
Experienced deep learning users who want to improve their understanding on how exactly Deep Neural network is able to come up with complex function approximations