Udemy
    •  
    •  
    •  
    •  
    •  
    •  
    •  
    •  
Subscribe
Turn what you know into an opportunity and reach millions around the world.
Learn More
Your cart is empty.
Keep shopping
Building a Neural Network from Zero
Rating: 4.3 out of 5(24 ratings)
7,341 students

Building a Neural Network from Zero

Master Neural Networks by Building from Scratch: Forward/Backward Pass, SGD, and Fashion-MNIST Challenge
Created byNick Ovchinnikov
Last updated 4/2025
English

What you'll learn

  • Implement neural networks from scratch, including forward and backward propagation
  • Master gradient descent, SGD with momentum, and other optimization techniques
  • Build custom layers, activation functions, and loss functions without external libraries
  • Apply your custom neural network to solve the Fashion-MNIST classification challenge

Course content

2 sections • 14 lectures • 3h 59m total length

  • Introduction0:19
  • Numerical Differentiation12:05
  • Numerical Differentiation: 3 ways14:44
  • Understanding Numerical Differentiation
  • Gradient Descend: 2d25:28
  • Gradient Descend: Multi-Dimensions18:43

    Explore the gradient as a vector of partial derivatives and how following the negative gradient enables gradient descent toward minima on 2D and 3D paraboloids, with learning rate and iterations.

  • SGD: Momentum20:57
  • Gradient Descent and Optimization
  • Cross-Entropy22:13
  • Sigmoid: Convert Logits to Probability12:14
  • Weights Init: He / Xavier17:51
  • Loss, Probs and Weights Init

Requirements

  • Basic knowledge of Python programming
  • Familiarity with linear algebra concepts like vectors and matrices
  • An interest in understanding neural networks at a fundamental level

Description

Are you ready to take your understanding of neural networks to the next level? In "Building a Neural Network from Zero," you'll dive deep into the inner workings of neural networks by implementing everything from scratch. This course is perfect for those who want to go beyond using libraries and truly understand how each component functions under the hood.

In this hands-on course, we will manually construct a PyTorch-like framework to build, train, and evaluate neural networks. Starting from the fundamentals of numerical differentiation and gradient descent, you'll gradually develop a complete training loop. You'll gain in-depth knowledge of essential concepts, including:

  • Numerical differentiation and three approaches to compute gradients

  • Gradient descent in 2D and multi-dimensional spaces

  • Stochastic Gradient Descent (SGD) with momentum

  • Implementing cross-entropy loss and activation functions like Sigmoid

  • Initializing neural network weights using He and Xavier methods

  • Building a fully functional Feedforward Neural Network (FFNN) from scratch

By the end of the course, you'll have a comprehensive understanding of how neural networks learn. To solidify your knowledge, we'll tackle the Fashion-MNIST challenge, where you'll apply your custom-built neural network to classify images accurately.

Whether you're an aspiring machine learning engineer or a curious programmer, this course equips you with the foundational knowledge and hands-on experience to build and customize neural networks from the ground up.

Enroll today and start mastering neural networks by building them from scratch!

Who this course is for:

  • Beginners who want to understand how neural networks work under the hood
  • Machine learning enthusiasts looking to deepen their knowledge through hands-on implementation
  • Developers who want to build custom neural network models from scratch
  • Students and professionals seeking to strengthen their grasp of core deep-learning concepts

Report abuse