Machine Learning: Neural networks from scratch

Implementation of neural networks from scratch (Python)

In this course, we will implement a neural network from scratch, without dedicated libraries. Although we will use the python programming language, at the end of this course, you will be able to implement a neural network in any programming language.

We will see how neural networks work intuitively, and then mathematically. We will also see some important tricks, which allow stabilizing the training of neural networks (log-sum-exp trick), and to prevent the memory used during training from growing exponentially (jacobian-vector product). Without these tricks, most neural networks could not be trained.

We will train our neural networks on real image classification and regression problems. To do so, we will implement different cost functions, as well as several activation functions.

This course is aimed at developers who would like to implement a neural network from scratch as well as those who want to understand how a neural network works from A to Z.

This course is taught using the Python programming language and requires basic programming skills. If you do not have the required background, I recommend that you brush up on your programming skills by taking a crash course in programming. It is also recommended that you have some knowledge of Algebra and Analysis to get the most out of this course.

Concepts covered :

• Neural networks

• Implementing neural networks from scratch

• Gradient descent and Jacobian matrix

• The creation of Modules that can be nested in order to create a complex neural architecture

• The log-sum-exp trick

• Jacobian vector product

• Activation functions (ReLU, Softmax, LogSoftmax, ...)

• Cost functions (MSELoss, NLLLoss, ...)

  
  

This course will be frequently updated, with the addition of bonuses.

Don't wait any longer before launching yourself into the world of machine learning!