In this module, you’ll explore logistic regression training and cross-entropy loss in PyTorch. You’ll examine why mean squared error performs poorly for classification and how maximum likelihood connects to cross-entropy loss. Additionally, you’ll explore loss behavior, optimization surfaces, and classification training loops. The module also enables you to practice these concepts through guided labs and quizzes that focus on PyTorch implementation patterns.

In this module, you’ll explore Softmax regression for multi-class classification and examine how Softmax converts model scores into class probabilities and how argmax supports prediction selection. You’ll practice building Softmax classifiers in PyTorch and step through end-to-end classification workflows. Further, you’ll implement Softmax-based models using PyTorch nn.Module patterns. Finally, you'll explore the role of activation functions in neural networks and learn about implementing Sigmoid, Tanh, and ReLU activation functions in PyTorch.

In this module, you’ll build and train shallow neural networks using PyTorch model patterns such as nn.Module and nn.Sequential. You’ll work with hidden layers, forward-pass computations, and activation functions to see how networks form non-linear decision boundaries. You’ll also construct networks for multi-dimensional inputs and multiclass classification tasks. The module enables examining how hidden neuron counts affect model capacity and training behavior. Finally, you’ll explore backpropagation, gradient flow, vanishing gradients, and the effects of overfitting and underfitting as you configure and adjust shallow network architectures.

In this module, you’ll construct deep neural networks using layered PyTorch architectures and flexible model patterns such as nn.ModuleList. You’ll configure multi-layer networks with different activation functions and layer sizes to examine how depth and structure affect training behavior. Further, you’ll apply techniques such as dropout, weight initialization methods, momentum-based optimization, and batch normalization to stabilize and accelerate training. Finally, you’ll explore how initialization choices and normalization layers influence gradient flow and convergence in deeper models.

In this module, you’ll build convolutional neural networks for image classification using PyTorch CNN components. You’ll apply convolution operations, stride, padding, activation maps, and pooling layers to understand how spatial features are detected and reduced across layers. Additionally, you’ll assemble CNN architectures and step through the constructor, forward pass, and training workflow in PyTorch. You’ll also learn to work with GPU and CUDA execution patterns and examine how hardware acceleration supports CNN training. Finally, you’ll explore residual network concepts, pretrained models such as ResNet18 with TorchVision, and transfer learning patterns used in modern CNN pipelines.

In this module, you’ll complete a guided final project focused on convolutional neural network classification in PyTorch. You’ll build, configure, and train a CNN using a structured dataset workflow and apply model setup, forward-pass, and training patterns. You’ll move through project design, model training, and evaluation steps as you assemble your solution.