Share Description
In this course, you will :
- Discover the inner workings of neurons and neural networks, as well as how activation functions, affine transformations, and layers interact within a deep learning model.
- Learn how a model like this is trained, or how the best values of model parameters are estimated.
- How smartly gradient descent optimization is used to optimise this process.
- Understand the various types of differentiation that could be used in this process, as well as how PyTorch implements reverse-mode auto-differentiation using Autograd.
- In PyTorch, learn how to create dynamic computation graphs.
Syllabus :
1. Getting Started with PyTorch for Machine Learning
- Representation Learning Using Neural Networks
- Neuron as a Mathematical Function
- Activation Functions
- Introducing PyTorch
- TensorFlow and PyTorch
- Demo: PyTorch Install and Setup
2. Working with Tensors in PyTorch
- Demo: Creating and Initializing Tensors
- Demo: Simple Operations on Tensors
- Demo: Elementwise and Matrix Operations on Tensors
- Demo: Converting between PyTorch Tensors and NumPy Arrays
- PyTorch Support for CUDA Devices
- Demo: Setting up a Deep Learning VM to Work with GPUs
- Demo: Creating Tensors on CUDA-enabled Devices
- Demo: Working with the Device Context Manager
3. Working with Gradients Using the Autograd Library
- Gradient Descent Optimization
- Forward and Backward Passes
- Calculating Gradients
- Using Gradients to Update Model Parameters
- Two Passes in Reverse Mode Automatic Differentiation
- Demo: Introducing Autograd
- Demo: Working with Gradients
- Demo: Variables and Tensors
- Demo: Training a Linear Model Using Autograd
4. Building Dynamic Computation Graphs
- Static vs. Dynamic Computation Graphs
- Dynamic Computation Graphs in PyTorch
- Demo: Installing Tensorflow, Graphviz, and Hidden Layer
- Demo: Building Dynamic Computations Graphs with PyTorch
- Demo: Visualizing Neural Networks in PyTorch Using Hidden Layer
- Demo: Building Static Computation Graphs with Tensorflow
- Demo: Visualizing Tensorflow Graphs with Tensorboard
- Demo: Dynamic Computation Graphs in Tensorflow with Eager Execution
- Debugging in PyTorch and Tensorflow
