02. Review: Training a Neural Network

Elective: Review and Learn PyTorch

For this course, you are expected to know how neural networks train through backpropagation and have an idea of what loss functions you can use to train a CNN for a classification task. If you’d like to review this material, you can look at the Elective Section, Review: Training a Neural Network (at the bottom of all the lessons on the main course page) and pay close attention to the videos in that section.

This review section covers:

• How neural networks train and update their weights through backpropagation
• How to construct models in PyTorch

So, if you'd like a brief overview of neural networks and deep learning or if the PyTorch framework is new to you, make sure to take a look at that section!

Why PyTorch?

We'll be using PyTorch throughout this program. PyTorch is definitely a newer framework, but it's fast and intuitive when compared to Tensorflow variables and sessions. PyTorch is designed to look and act a lot like normal Python code: PyTorch neural nets have their layers and feedforward behavior defined in a class. defining a network in a class means that you can instantiate multiple networks, dynamically change the structure of a model, and these class functions are called during training and testing.

PyTorch is also great for testing different model architectures, which is highly encouraged in this course! PyTorch networks are modular, which makes it easy to change a single layer in a network or modify the loss function and see the effect on training. If you'd like to see a review of PyTorch vs. TensorFlow, I recommend this blog post.