Skip to Content
View all events

Deep Learning from Scratch with PyTorch

Published by O'Reilly Media, Inc.

Intermediate content levelIntermediate

Build a Neural Network with NumPy!

Course Outcomes:

  • Understand the basics of neural networks and deep learning models, a building block for computer vision and large language models
  • Build a neural network completely from scratch in NumPy, including gradient descent and backpropagation
  • Perform train/test splits and evaluate performance, and compare your “from scratch” models to PyTorch

Course description

Neural networks and deep learning have captured attention for years, and understandably so. Problems that were previously hard for computers to solve like language, computer vision, audio recognition have made enormous strides due to these powerful techniques. Even large large language models and agentic AI have a deep learning component. But how do these mysterious black boxes work? Join expert Thomas Nield to learn how to build a neural network completely from scratch using NumPy that will predict a light or dark font for a given background color. Whether you are beginner or advanced, by understanding the “learning” behind deep learning, you will have a solid intuition and building block on what makes modern AI work. You will also get a solid foundation for more advanced applications, like computer vision and large language models. You’ll pick up best practices, such as train/test splits and understand how hyperparameters (like learning rate) affect results.

What you’ll learn and how you can apply it

  • Build Neural Networks: Use PyTorch to construct and execute a basic deep learning model from the ground up.
  • Mathematical Troubleshooting: Apply Linear Algebra and Calculus (derivatives) to understand why a model is failing or how to optimize its learning speed.
  • Data Processing: Manually implement Forward Propagation to pass raw data through a network and generate predictions.
  • Model Training: Use Gradient Descent and Backpropagation to "teach" a model by calculating error and updating internal weights.
  • Image Recognition: Build a functional Computer Vision classifier using the MNIST dataset to identify handwritten digits.
  • AI Architecture Analysis: Evaluate the mechanics behind high-level tech like LLMs (Large Language Models) and autonomous systems (Self-driving cars).
  • Performance Auditing: Use standard metrics to determine if a model is production-ready or needs further training.

This live event is for you because...

  • You’re a budding data science professional or software engineer seeking to understand neural networks and deep learning.
  • You’re a data science professional or software engineer who wants to understand the use cases, context, and appropriate practices for implementing a deep learning system.
  • You have some Python proficiency and want to learn about neural networks and deep learning.
  • You’re a project manager who is being asked to apply deep learning to a business.

Prerequisites

Recommended follow-up:

Schedule

The time frames are only estimates and may vary according to how the class is progressing.

Introduction (5 minutes)

Neural Networks and Deep Learning Basics (45 mins)

  • What are neural networks?
  • What is deep learning?
  • Large language models and computer vision
  • LAB: A basic example in PyTorch
  • EXERCISE: Complete the PyTorch code

BREAK: 10 minutes

Forward Propagation (50 mins)

  • What is forward propagation?
  • Linear algebra crash course
  • Passing a color through a neural network
  • LAB: Forward propagation in Python
  • EXERCISE: Complete the forward propagation code

BREAK: 10 minutes

Gradient Descent (50 mins)

  • What are derivatives?
  • Using SymPy to calculate derivatives
  • What is gradient descent?
  • LAB: A simple gradient descent problem
  • LAB: Gradient descent with linear regression
  • What is stochastic gradient descent?
  • EXERCISE: Complete the gradient descent code

BREAK: 10 minutes

Backpropagation (50 mins)

  • The Chain Rule
  • Using the chain rule for backpropagation
  • LAB: Implementing backpropagation
  • EXERCISE: Complete the backpropagation code

BREAK: 10 minutes

Other Applications and Production (45 mins)

  • Evaluating performance
  • Variants of neural networks
  • Basic computer vision with the MNIST dataset
  • Relating to generative AI and large language models
  • CASE STUDY: How do self-driving cars work?

Closing (5 minutes)

Your Instructor

  • Thomas Nield

    Thomas Nield is the founder of Nield Consulting Group and an instructor at University of Southern California, where he teaches AI System Safety, developing systematic approaches for identifying AI-related hazards in aviation and ground vehicles. He's authored three books, including Essential Math for Data Science and Getting Started with SQL (both for O'Reilly). He enjoys making technical content relatable and relevant to those unfamiliar or intimidated by it. Thomas teaches classes on data analysis, machine learning, mathematical optimization, and practical artificial intelligence. He’s also the founder and inventor of Yawman Flight, a company that develops universal handheld flight controls for flight simulation and unmanned aerial vehicles.

    Xlinksearch

Skills covered

  • Deep Learning
  • OpenAI
  • Pandas
  • Reinforcement Learning