Introduction to Neural Networks and PyTorch

You will work with 1D and 2D tensor operations, PyTorch autograd and automatic differentiation, linear regression with nn.Module and custom modules, MSE and cross-entropy loss functions, batch/stochastic/mini-batch gradient descent, DataLoader and optimizer workflows, training/validation/test splits, logistic regression with sigmoid thresholding, and advanced optimization techniques including Adam, learning rate scheduling, and L1/L2 regularization.

In the final project, you will build a logistic regression model to predict League of Legends match outcomes using real in-game statistics—applying the complete PyTorch workflow from data preparation through model training and evaluation. The project produces a portfolio-ready deliverable that demonstrates your ability to implement a classification pipeline end to end.

This course establishes the underlying mathematical workflows of modern AI engineering. Instead of manually calculating derivatives for backpropagation, you will master PyTorch's autograd system. You will learn how one- and two-dimensional tensors track their computational history, enabling automatic differentiation. By understanding attributes like shape, dtype, and numel(), you will see how the autograd.backward() function calculates vector-matrix derivatives behind the scenes to dynamically update model parameters.

Writing production-grade machine learning code requires leveraging PyTorch’s native data abstraction utilities. The curriculum provides hands-on practice building structured, memory-efficient data pipelines by sub-classing core PyTorch components:

Dataset Class: Mastering how to override __len__ and __getitem__ to read custom tabular or image files alongside Torchvision Compose data transforms.

DataLoader Class: Implementing multi-threaded data streaming to handle batching, shuffling, and mini-batch gradient descent convergence.

nn.Module Architecture: Packaging weights, biases, and layers into an object-oriented model wrapper while leveraging the state_dict() to save, load, and transfer learned parameters.

A key analytical threshold in this specialization is migrating from regression (predicting continuous variables) to classification (predicting discrete labels). You will learn how to expand custom layers using nn.Linear and nn.Sequential. The course breaks down the probabilistic shift under the hood, teaching you how to pass linear combinations through a sigmoid activation function to model a Bernoulli distribution. Crucially, you will analyze why Binary Cross-Entropy Loss and Maximum Likelihood Estimation are mathematically superior to Mean Squared Error (MSE) when training classification networks.

To access the course materials, assignments and to earn a Certificate, you will need to purchase the Certificate experience when you enroll in a course. You can try a Free Trial instead, or apply for Financial Aid. The course may offer 'Full Course, No Certificate' instead. This option lets you see all course materials, submit required assessments, and get a final grade. This also means that you will not be able to purchase a Certificate experience.