What PyTorch architectures are covered in this deep learning masterclass?

The course covers CNNs, ResNets, autoencoders, VAEs, GANs, U-Nets, RNNs, LSTMs, attention mechanisms, Transformers, Vision Transformers, and diffusion models.

How does the course explain PyTorch training loops and loss functions?

It demonstrates the standard training workflow—forward pass, loss calculation, gradient reset, backpropagation, and optimizer update—along with losses such as CrossEntropyLoss and BCEWithLogitsLoss.

Does the course include practical PyTorch projects beyond image classification?

Yes. Projects include object detection, image segmentation, text classification and generation, image captioning, semantic clustering, sentiment analysis, ONNX export, and Raspberry Pi deployment.

Free Course Image PyTorch Deep Learning Masterclass: CNNs, Transformers, GANs, Autoencoders and Diffusion

Free online course PyTorch Deep Learning Masterclass: CNNs, Transformers, GANs, Autoencoders and Diffusion

Duration of the online course: 18 hours and 13 minutes

New

Build real-world deep learning skills with a free PyTorch course—train CNNs, Transformers, GANs and diffusion models, plus projects and a shareable certificate.

In this free course, learn about

How PyTorch tensors, shapes, and batch dimensions work (e.g., unsqueeze for batching)
Core autograd/optimization concepts, including why updates use torch.no_grad()
The standard 5-step PyTorch training loop and how to structure training/validation/test
Binary vs multiclass losses: BCEWithLogitsLoss, CrossEntropyLoss, and target/output formats
Activation functions and why nonlinearity is required between linear layers
MLPs for regression/classification and linear classification fundamentals
CNN basics: conv kernels/weight shapes, LeNet-style MNIST, ResNets, and transfer learning
Data augmentation and self-supervised CNN pretraining (e.g., rotation prediction labels)
Autoencoders, VAEs, VQ-VAEs: latent bottlenecks, stochastic latents, and codebook quantization
Object detection and U-Net segmentation, including skip-connection purpose
Generative models: GAN discriminator objective and diffusion forward noising process
Sequence models: RNN/LSTM design choices, vocab building, seq2seq, and attention mechanics
Transformers for text/vision: positional embeddings, causal masking, cross-attention, patches
Deployment/production: ONNX export needs example inputs; running models on Raspberry Pi

About the free online course

Turn PyTorch into a practical skill you can use to build modern AI systems end to end. This free online course is designed for learners who want more than high-level theory: you will develop intuition for how tensors, gradients and training loops work, then apply that foundation to create models that solve real problems in computer vision and natural language processing.

You’ll start by strengthening core deep learning habits in PyTorch—shaping data correctly, organizing experiments, and understanding what happens during forward and backward passes. From there, you’ll progress into model design choices that directly impact performance: activation functions, loss functions, optimization patterns, and the small implementation details that often decide whether training is stable or frustrating. Along the way, the course emphasizes the reasoning behind common PyTorch practices so you can debug with confidence rather than guessing.

As you advance, you’ll build capable vision models using convolutional networks, learn how deeper architectures such as residual networks unlock better feature learning, and see how transfer learning speeds up results when data is limited. You will also explore techniques that make models generalize better, including data augmentation and strategies for learning from unlabeled data. That practical focus continues with architectures used in applied settings, such as object detection and image segmentation, connecting research ideas to workflows that resemble what teams use in production.

The course then expands into generative modeling, where you’ll learn how autoencoders and variational autoencoders compress and reconstruct data, how GANs learn to synthesize realistic samples, and how diffusion models generate images through iterative denoising. You will also build sequence models, moving from recurrent networks and LSTMs to attention mechanisms and Transformers, gaining a clear understanding of why self-attention, positional embeddings, and masking are essential for language tasks.

To round out your skill set, you’ll work with interpretability concepts, semantic embeddings, and practical deployments such as exporting models and running inference on constrained hardware. By the end, you will have a coherent PyTorch toolkit for training, evaluating, and shipping neural networks across vision, text, and generative AI—ready for portfolios, interviews, and real product work.

Course content

Video class: Introduction to the Pytorch Deep Learning Tutorial Series 02m
Exercise: How is each section of the course repository organized?
Video class: Pytorch Deep Learning Tutorial: Deep Dive into Basics (Part 1) 28m
Exercise: In PyTorch, why is adding a leading batch dimension (often via unsqueeze(0)) commonly needed?
Video class: Pytorch Deep Learning Tutorial: Building Strong Foundations (Part 2) 22m
Exercise: When manually updating a tensor parameter during gradient descent in PyTorch, why is the update typically wrapped in torch.no_grad()?
Video class: Pytorch Deep Learning Tutorial: Advanced Techniques Unveiled (Part 3) 22m
Exercise: Which sequence matches the typical 5-step PyTorch training loop for one iteration?

Video class: Performing Neural Network Linear Classification: Pytorch Deep Learning Tutorial 36m
Exercise: Why is BCEWithLogitsLoss often preferred over using a sigmoid output with MSELoss for binary classification?
Video class: Understanding Neural Network Activation Functions: Pytorch Deep Learning Tutorial 28m
Exercise: Why are non-linear activation functions (e.g., sigmoid) needed between linear layers in a neural network?
Video class: Multilayer Perceptron Regression: Pytorch Deep Learning Tutorial 35m
Video class: Multilayer Perceptron Classification: Pytorch Deep Learning Tutorial 36m
Exercise: When using PyTorch's CrossEntropyLoss for MNIST classification, what should the model output and target labels look like?

Video class: What are Convolution layers? : Pytorch Deep Learning Tutorial 37m
Exercise: In PyTorch, what kernel tensor shape is expected for a 2D convolution weight when manually using a convolution function?
Video class: Performing Classification with a CNN (Start HERE!): Pytorch Deep Learning Tutorial 33m
Exercise: In a basic LeNet-style CNN for MNIST, why is a validation split created in addition to the test set?
Video class: Creating DEEP CNNs with ResNets: Pytorch Deep Learning Tutorial 42m
Video class: Using Transfer Learning With Neural Networks: Pytorch Deep Learning Tutorial 31m
Exercise: In transfer learning with a pre-trained ResNet-18 on a new 10-class dataset, what is the typical best practice for adapting the model output?

Video class: Pytorch Data Augmentation for CNNs: Pytorch Deep Learning Tutorial 39m
Exercise: What is the main purpose of applying data augmentation during training for an image classifier?
Video class: Unsupervised Learning Strategies for a CNN: Pytorch Deep Learning Tutorial 23m
Exercise: In rotation-based pre-training, what target label does the model learn to predict for each image?
Video class: Creating and Training Autoencoders: Pytorch Deep Learning Tutorial 44m
Exercise: In an autoencoder, what is the main purpose of the bottleneck (latent code) between the encoder and decoder?
Video class: Creating and Training Variational Autoencoders: Pytorch Deep Learning Tutorial 26m
Exercise: In a Variational Autoencoder (VAE), what key change is made to the bottleneck compared to a vanilla autoencoder?

Video class: Simple Object Detection with a CNN (From Scratch): Pytorch Deep Learning Tutorial 38m
Video class: U-net Image Segmentation the basics (From Scratch!): Pytorch Deep Learning Tutorial 28m
Exercise: In a U-Net used for image segmentation, what is the main purpose of the skip connections between encoder and decoder layers?
Video class: Generative Adversarial Networks (GANs) the basics from scratch! : PyTorch Deep Learning Tutorial 32m
Exercise: In a GAN training setup, what is the discriminator primarily learning to decide?
Video class: Generative Image Diffusion from Scratch! PyTorch Deep Learning Tutorial 44m
Exercise: In diffusion-based image generation, what is the forward process?

Video class: Local Interpretable Model-agnostic Explanations (LIME): PyTorch Deep Learning Tutorial 36m
Exercise: In LIME, how is a complex neural network explained for a single input instance?
Video class: Training Autoregressive MLPs, the basics! PyTorch Deep Learning Tutorial 22m

Video class: Creating and Training Recurrent Neural Networks! : PyTorch Deep Learning Tutorial 23m
Exercise: Why is a tanh activation used when producing the recurrent summary/buffer vector?
Video class: Long-Short Term Memory and how to use it! PyTorch Deep Learning Tutorials 42m
Exercise: What key design choice in an LSTM helps reduce the vanishing-gradient problem compared with a basic RNN?
Video class: Creating a Text Classifier With LSTM! PyTorch Deep Learning Tutorial 32m
Exercise: In an LSTM text classification pipeline, what is the main purpose of building a vocabulary (vocab) after tokenizing the sentences?
Video class: Building a Text Generator With LSTM, From Scratch!: PyTorch Deep Learning Tutorial 30m
Exercise: In next-token prediction training for an LSTM text generator, how are the input and target sequences typically formed from a tokenized sentence?
Video class: Sequence to Sequence Processing with LSTMs, From Scratch: PyTorch Deep Learning Tutorial 15m
Exercise: In a sequence-to-sequence LSTM question-answering setup, when is the training loss calculated?

Video class: Implementing the Attention Mechanism from scratch: PyTorch Deep Learning Tutorial 47m
Exercise: In the attention mechanism described, why is a softmax applied to the attention scores before multiplying by the values (V)?
Video class: Let's Add Attention to a LSTM Network! PyTorch Deep Learning Tutorial 18m
Exercise: In an LSTM enhanced with attention for text generation, what is stored and later used as the keys and values for the attention mechanism?
Video class: Adding Self-Attention to a Convolutional Neural Network! PyTorch Deep Learning Tutorial 14m
Exercise: In a CNN with added self-attention, how is a 2D feature map typically prepared so it can be fed into PyTorch multi-head attention?

Video class: Classifying Text with a Transformer LLM, From Scratch! : PyTorch Deep Learning Tutorial 18m
Exercise: Why are positional embeddings needed in a Transformer encoder for text?
Video class: Decoder-Only Transformer for Next Token Prediction: PyTorch Deep Learning Tutorial 15m
Exercise: In a decoder-only Transformer for next-token prediction, what is the main purpose of causal masking in self-attention?
Video class: Building an Encoder-Decoder Transformer from Scratch!: PyTorch Deep Learning Tutorial 15m
Exercise: In an encoder-decoder Transformer, what is the main purpose of the decoder’s cross-attention layer?
Video class: Classify Images with a Vision Transformer (ViT): PyTorch Deep Learning Tutorial 10m
Exercise: In a Vision Transformer for image classification, what is the main reason for splitting an image into patches before applying the Transformer encoder?

Video class: Deploying Pytorch Neutral Networks On the Raspberry Pi! 11m
Exercise: When exporting a PyTorch model to ONNX, why is an example input tensor required?
Video class: Captioning Images with a Transformer, from Scratch! PyTorch Deep Learning Tutorial 18m
Exercise: In a Transformer-based image captioning model, what is the main purpose of using a vision encoder together with a text decoder?

Video class: Latent Diffusion for Image Generation with a Unet: PyTorch Deep Tutorial 13m
Exercise: What is the key difference between latent diffusion and image diffusion?
Video class: Diffusion with Transformers AND Diffusion In-Painting from Scratch! PyTorch Deep Tutorial 20m
Exercise: In a diffusion transformer for latent diffusion, what is the key architectural change compared to a vision transformer used for image classification?

Video class: Semantic Clustering of Text using pre-trained HuggingFace models! 12m
Exercise: In semantic clustering with a pre-trained text transformer, what does it mean when two reviews have embeddings that are close together in the embedding space?
Video class: Stock Market Sentiment Analysis with LLMs! 09m
Exercise: How is the daily sentiment score constructed from the model’s outputs?

Video class: Creating a Vector Quantized VAE from Scratch! PyTorch Deep Tutorial 29m
Exercise: In a Vector-Quantized VAE (VQ-VAE), how is an encoder output vector chunk mapped to the codebook during quantization?

Learn aboutPython

Explore free online Python courses and build in-demand programming skills at your own pace. Learn Python basics, data analysis, automation, web development, and more through flexible beginner-friendly lessons. Enroll today, study from anywhere, and earn a free certificate to showcase your Python skills and boost your career.

Learn aboutDeep Learning

Explore the best free online Deep Learning courses and build practical skills in neural networks, CNNs, RNNs, transformers, computer vision, and AI. Learn at your own pace with expert-led lessons, hands-on projects, and real-world examples. Every course is free and includes a certificate to help you advance your career in machine learning and artificial intelligence.

Learn aboutPyTorch

Discover free online PyTorch courses that include a certificate and learn to build, train, and deploy deep learning models. Master tensors, neural networks, computer vision, natural language processing, and GPU acceleration through flexible, beginner-friendly lessons. Start learning PyTorch today, strengthen your AI skills, and earn a certificate at no cost.