Deep learning models are transforming society. Everyone can understand how these models work. This course walks through all the important concepts relating to deep learning models and how they give rise to the recent results in AI. We use guided examples and discuss a variety of practical applications, all accompanied by animations and visualizations. We also cover recent breakthroughs in deep learning research. This course will demystify the models that underpin the recent AI revolution and provide a solid foundation for further learning.
What am I going to get from this course?
After taking this course students will be able to understand:
- What deep learning is and how it differs from other types of machine learning and artificial intelligence
- How deep learning models use neural networks to make computations
- What types of problems deep learning models can be used to solve
- The types of data needed to train deep learning models
- The variety of inputs deep learning models receive and solutions they produce
- The advantages that deep learning can offer over traditional machine learning
- Why multi-neuron networks are able to solve complex problems
- How neural networks use gradient descent and back-propagation to learn to make predictions
- All material will be accompanied with ample visualisations and animations to provide intuitive understanding
Prerequisites and Target Audience
What will students need to know or do before starting this course?
No prerequisites required.
Who should take this course? Who should not?
Who should take Introduction to Deep Learning:
- Working Professionals who work around AI and technology and would like to have a clearer understanding of how deep learning works
- Managers and Executives from any field who would like to have a firmer grasp of the foundations of cutting-edge Artificial Intelligence
- Data scientists and machine learning practitioners who would like to expand their knowledge to deep learning
- Aspiring deep learning practitioners who want to an introduction that provides friendly examples and intuition while still covering the background needed to enables further learning for serious deep learning work.
- Students and fresh graduates who are interested in a career working around or in Data Science Machine Learning and Artificial Intelligence
- Everyone who would like to understand the fundamental principles that underlie cutting-edge Artificial Intelligence
Who should not take Introduction to Deep Learning:
You want software code-based examples of implementing deep learning models
Intelligence and Information Processing
In this lecture you will learn what information processing is and how intelligence relates to information processing. You will learn how artificial intelligence is achieved through models that solve information processing problems. (Transcript attached)
Artificial Intelligence, Machine Learning and Deep Learning part 1
In this lecture go through examples of how AI solves complex information processing problems. You will learn how deep learning sits within the larger context of artificial intelligence and traditional machine learning. (Transcript attached)
Artificial Intelligence, Machine Learning and Deep Learning part 2
In this lecture we go over an example of how deep learning differs from traditional AI and traditional machine learning. You will also learn about advantages offered by deep learning over traditional machine learning and what factors contributed to the deep learning revolution.
Inputs, Outputs, & Features
In this lecture we go through example inputs deep learning models might receive and example output they might produce.
Importance of Feature Representations
In this lecture we go over why it is important that deep learning models can find efficient feature representations is important and how this relates to the function of our brains.
Efficient Feature Representation Example
Here we walk through an example of efficient feature representation.
Overview of supervised learning models
Overview of unsupervised learning models
Overview of supervised learning models
Module 2: Perceptron: Weights, Biases, Activation Functions
Introduction to Single Neuron Model
Introduction to single neuron models, which is the basic computing component of deep learning models.
Single Neuron Example: 1D Example of Bluffing at Poker
Walk through of an example of a simple neuron model solving the problem of predicting whether your friend is bluffing at poker. This will provide the groundwork for understanding how neural network models make predictions from data.
Single Neuron Example: 2D Example of Bluffing at Poker
A further example of a single neuron model making predictions, applied to two dimensional data.
Single Neuron Model Limitations
Discussion of the limitations of a single neuron model and the inability of single neurons to solve nonlinearly separable problems.
Module 3: Multi-neuron Networks : XOR and nonlinearity
Binary Logical Operators and XOR
Explanation of XOR, which is an example of a nonlinearly separable problem that a single neuron will not be able to solve.
Multi-neuron solution to XOR
Walk-through of how a multi-neuron network can solve the XOR problem.
Universal Approximation Theorem
A visual sketch of the proof of the universal approximation theorem, which shows how a multi-neuron network can represent any nonlinear function.
Neural Network Walk-through
We step through how a fully connected feed forward neural network takes in input and produces an output.
Module 4: Learning: Gradient Descent
In this lecture you will learn about the cost function and how the reduction of the cost function is the goal of neural network learning.
Derivative and Gradient Part 1
In this lecture you will learn about derivatives and gradients, which tell us how to change parameter values in order to minimise the cost function. You will also learn how the gradient is used to update neural network parameters reduces the cost function as the neural network learns.
Derivative and Gradient Part 2
More about derivatives and gradients
Neural Network Parameters and Matrix Notation
In this lecture we will walk through the computations of a fully connected feedforward neural network. We will count the parameters that need to be learned. You will also learn linear algebra notation.
In this lecture we cover the chain rule and how neural networks use the chain rule to calculate the gradient for learning.
Backpropagation and Overfitting
<div><!--block-->In this lecture we cover backpropagation, the algorithm neural networks use to learn. You will also learn about the problem of overfitting.<br> (Transcript attached)<br><br></div>
Backpropagation for Handwritten Digit Recognition
<div><!--block-->Example of how backpropagation is used to train a neural network to recognise handwritten digits.<br><br> (Transcript attached)</div>
Variations on Learning
<div><!--block-->In this lecture we cover variations on backpropagation and discuss additional considerations when training neural networks.<br><br> (Transcript attached)<br><br></div>