Machine Learning, Deep Learning & Neural Networks in Matlab

Name: Machine Learning, Deep Learning & Neural Networks in Matlab
Rating: 4.2 (163 reviews)

Learn deep learning from A to Z and create a neural network in MATLAB to recognize handwritten numbers (MNIST database)

Created byEliott Wertheimer, Albert Nassar

Last updated 4/2020

English

English [Auto],

What you'll learn

How neural networks emulate the brain
How to artificially represent neural networks
The biological fundamentals behind neural networks and deep learning
How to represent and manipulate neural networks with matrices
The basics behind training neural networks and the cost function
How to use gradient descent and learning intuition
The maths and calculus behind forward and back propagation
How to represent forward and back propagation in matrix form
How to write forward and back propagation algorithms
Understand and master the mathematics and algorithms behind deep learning and neural networks
The structure of the MNIST database and how to use and extract data from it
How to program and use the Sigmoid and Leaky Relu activation functions in MATLAB
How to create a Neural Network in Matlab
How to create Neural Network training and testing algorithms in Matlab
How to use the MNIST database to make a neural network able to read handwritten numbers in images

Course content

2 sections • 22 lectures • 4h 4m total length

Introduction to Deep Learning and Neural Networks4:22
Explore the fundamentals of deep learning and neural networks in Matlab, including theory, mathematics, and how networks learn. Apply practical and algorithmic approaches to build and evaluate neural networks.
Emulating the brain6:45
Explore how biology informs artificial neural networks, from neurons and synapses to input, hidden, and output layers, and discuss the universal approximation theorem and learning limits.
Artificial Neural Network Representation7:51
Explore how artificial neural networks represent data through neurons, weights, biases, and activation functions across input, hidden, and output layers, and how nonlinear, differentiable activations enable learning.
Matrix Representation of Neural Networks4:33
Explore how neural networks use matrix representations to express inputs, weights, biases, and activation functions, enabling efficient dot-product computations across hidden layers to outputs.
Nomenclature and Notations1:52
Explore neural network nomenclature and notations, including layer and node indices, weights and biases, pre-activation inputs, activations, the loss function, and the backpropagation framework.
Training Basics and the Cost Function4:12
Explore how learning uses the cost function to measure error and guide training, adjusting weights and biases to minimize the average squared error over a dataset.
Gradient Descent and Learning Intuition4:00
Explore gradient descent to minimize the cost function by computing gradients and updating weights and biases in the opposite direction. Learn how backpropagation drives gradients toward minima and global optima.
Backpropagation Calculus and Mathematics8:46
Explain back propagation calculus and the gradient of the cost function using chain rule and delta terms to compute derivatives of weights and biases in a simple sigmoid network.
Backpropagation Algorithm5:53
Explore backpropagation in a multi-neuron network with two input and two output neurons, using a sum of squared errors cost and delta terms to compute gradients for weights and biases.
Backpropagation in Matrix Form5:37
Convert backpropagation to matrix form to compute gradients of weights and biases, using forward pass to obtain A and Z and delta terms via activation derivatives.
Algorithm Implementation7:40
Apply forward propagation to compute z and l, then backpropagate deltas, average gradients over m samples, and update weights and biases via gradient descent with learning rate alpha.
Putting it All Together4:56
Putting it all together builds a neural network with learning via gradient descent and backpropagation, addresses data requirements, overfitting, hyperparameters, and the black box challenge, plus practical implementation.

The MNIST Handwritten Numbers Database9:08
Learn to work with the MNIST handwritten digit dataset in MATLAB, using 60k training and 10k test 28x28 grayscale images and IDX files to train a neural network.
Extracting the Images from the Database into MATLAB18:20
Develop a MATLAB read_images function to load binary image data, parse the header (magic number 2051, counts, rows, columns), and produce a 3D image matrix for training and testing.
Extracting the Relevant Image Labels from the MNIST Database into MATLAB8:14
Extract and read MNIST labels in MATLAB by implementing a read labels function that parses the binary label file, validates the header, and aligns resulting labels with the corresponding images.
The Sigmoid and Leaky Relu Activation Functions13:09
Create a Matlab script for activation functions, implementing sigmoid and leaky relu with an optional derivative, to activate neural network nodes across multiple inputs.
Initializing the Neural Network29:35
Initialize and train a Matlab neural network by setting up weights and biases, applying forward and backward propagation, and using adaptive learning with normalized inputs.
Neural Network Forward Propagation24:37
Implement forward propagation for a Matlab neural network, initialize gradients, weights, and biases, and train over 28x28 images with a loop that tracks error and accuracy.
Neural Network Backwards Propagation20:40
Learn to compute gradients for weights and biases using backpropagation in MATLAB, reducing the neural network's cost by updating parameters after forward propagation and analyzing hidden layers.
Neural Network Training16:22
Train a neural network in Matlab by computing squared error, tracking accuracy via the output's max, and updating weights and biases with gradient descent and an adaptive learning rate.
Creating a Testing Forward Propagation Algorithm15:28
Create a MATLAB testing script that runs forward propagation on the test set using training weights. Compute the predictions and the test success rate with a two-hidden-layer network.
Training and Testing our Neural Network using the MNIST database22:49
Train and test a Matlab-based neural network on the MNIST database, using two hidden layers of 20 nodes and 10 outputs, iterating debugging and training to achieve high test accuracy.

Requirements

High school mathematics level
Very basic MATLAB programming knowledge

Description

AI is omnipresent in our modern world. It is in your phone, in your laptop, in your car, in your fridge and other devices you would not dare to think of. After thousands of years of evolution, humanity has managed to create machines that can conduct specific intelligent tasks when trained properly. How? Through a process called machine learning or deep learning, by mimicking the behaviour of biological neurons through electronics and computer science. Even more than it is our present, it is our future, the key to unlocking exponential technological development and leading our societies through wonderful advancements.

As amazing as it sounds, it is not off limits to you, to the contrary!

We are both engineers, currently designing and marketing advanced ultra light electric vehicles. Albert is a Mechanical engineer specializing in advanced robotics and Eliott is an Aerospace Engineer specializing in advanced space systems with past projects completed in partnership with the European Space Agency.

The aim of this course is to teach you how to fully, and intuitively understand neural networks, from their very fundamentals. We will start from their biological inspiration through their mathematics to go all the way to creating, training and testing your own neural network on the famous MNIST database.

It is important to note that this course aims at giving you a complete and rich understanding of neural networks and AI, in order to give you the tools to create your own neural networks, whatever the project or application. We do this by taking you through the theory to then apply it on a very hands-on MATLAB project, the goal being for you to beat our own neural network's performance!

This course will give you the opportunity to understand, use and create:

How to emulate real brains with neural networks.
How to represent and annotate neural networks.
How to build and compute neural networks with matrices.
Understand and master the mathematics and algorithms behind deep learning and neural networks.
Train and test neural networks on any data set.
How to use the MNIST handwritting numbers training and testing datasets.
Import the MNIST data in MATLAB.
Create a complete neural network in MATLAB including forward and backwards propagation with both Leaky Relu and Sigmoid activation functions.
Train and test your own neural network on the MNIST database and beat our results (95% success rate).

We will thoroughly detail and walk you through each of these concepts and techniques and explain down to their fundamental principles, all concepts and subject-specific vocabulary. This course is the ideal beginner, intermediate or advanced learning platform for deep learning and neural networks, from their fundamentals to their practical, hands-on application. Whatever your background, whether you are a student, an engineer, a sci-fi addict, an amateur roboticist, a drone builder, a computer scientist, a business or sports person or anyone with an interest in data science and machine learning, at the end of this course, you will be capable of creating brains within machines!

If you have questions at any point of your progress along the course, do not hesitate to contact us, it will be our pleasure to answer you within 24 hours!

If this sounds like it might interest you, for your personal growth, career or academic endeavours, we strongly encourage you to join! You won't regret it!

Who this course is for:

Anyone interested in Artificial Intelligence
Anyone interested in Machine Learning
Anyone interested in Deep Learning
Annyone interested in Neural Networks
Anyone who wants to learn MATLAB while applying it to Deep Learning and Neural Networks
Anyone interested in creating Artificial Intelligence able to recognize handwritten numbers
Anyone interested in using and understanding the MNIST database to train in Neural Networks
Anyone interested to expand his knowledge in Data Science within his current career or as a new career
Anyone interested in entering the Data Science industry as a developer or an entrepreneur
Anyone who want to create value in their projects or businesses bu deeply understanding and leveraging data science and deep learning

Machine Learning, Deep Learning & Neural Networks in Matlab

What you'll learn

Explore related topics

Course content

The Theory: From its Biological Inspiration to Deep Learning12 lectures • 1hr 6min

MATLAB: create, train and test your neural network with the MNIST database10 lectures • 2hr 58min

Requirements

Description

Who this course is for: