Deep Learning with PyTorch – Workshop

Overview 📋

This hands-on workshop introduces the fundamentals of deep learning using PyTorch. Participants will learn by building real models and solving practical tasks across two sessions at Nancy Rothwell – 2A.011 M&T, on 23 February and 2 March 2025.

What You'll Learn

• Core PyTorch concepts (tensors, autograd, GPU usage)
• Building and training artificial neural networks and autoencoders
• Implementing CNNs for image classification tasks
• Applying transfer learning with pre-trained models for image segmentation
• Working with real-world datasets for classification, regression, and anomaly detection
• Understanding data preprocessing, augmentation, and normalisation techniques

---

Getting Started 🛠️

✅ Recommended Platform: Google Colab

Colab provides a free, GPU-enabled environment and is the primary platform for this workshop. No local installation is required.

What You Need

• A Google account
• Reliable internet connection

Running the Notebooks on Colab

1. Open the GitHub repo and click the "Open in Colab" badge at the top of any notebook, or:
   • Download the notebook locally.
   • Open Google Colab.
   • Use File > Upload Notebook to load it.
2. Enable GPU: Runtime > Change runtime type > Hardware Accelerator > GPU
3. Run the first setup cell to install all required dependencies.

📘 Colab Tips | Colab FAQ

---

💻 Running Locally (Optional)

Note: Local setup is optional. Google Colab is strongly preferred for the workshop.

Requirements

• Python 3.14+
• pip or uv (recommended for faster installs)

Setup Steps

Option A – using uv (recommended):

# 1. Install uv (if not already installed)
pip install uv

# 2. Clone the repository
git clone https://github.com/CLDiego/uom_fse_dl_workshop.git
cd uom_fse_dl_workshop

# 3. Create a virtual environment and install all dependencies
uv sync

# 4. Launch Jupyter
uv run jupyter notebook

Option B – using pip:

# 1. Clone the repository
git clone https://github.com/CLDiego/uom_fse_dl_workshop.git
cd uom_fse_dl_workshop

# 2. Create and activate a virtual environment
python -m venv .venv
source .venv/bin/activate        # macOS/Linux
# .venv\Scripts\activate         # Windows

# 3. Install the project and all dependencies (from pyproject.toml)
pip install .

# 4. Launch Jupyter
jupyter notebook

Then open any SE##_CA_*.ipynb notebook and run the first setup cell.

---

Workshop Schedule 🧠

Day 1 – 23 February (Sessions 1–3)

Time	Activity	Notebook
08:30	Registration, setup & troubleshooting
09:00 – 10:15	Session 1: Introduction to PyTorch
10:15 – 10:30	Break
10:30 – 11:45	Session 2: Artificial Neural Networks
11:45 – 12:45	Lunch
12:45 – 14:00	Session 3: Training Neural Networks

Day 2 – 2 March (Sessions 4–5)

Time	Activity	Notebook
08:30	Setup continuation
09:00 – 10:30	Session 4: Convolutional Neural Networks (CNNs)
10:30 – 10:45	Break
10:45 – 12:00	Session 5: Transfer Learning & Image Segmentation
12:00 – 13:00	Lunch
13:00 – 14:00	Wrap-up, discussion & Q&A

Light refreshments and lunch will be provided on both days.

---

Learning Outcomes 🎯

By the end of the workshop, you'll be able to:

• Implement deep learning models using PyTorch
• Build and train artificial neural networks and autoencoders
• Apply CNNs for image classification tasks
• Use transfer learning with pre-trained models for image segmentation
• Work with real-world datasets for classification, regression, and anomaly detection tasks
• Understand data preprocessing, augmentation, and normalisation techniques

---

Datasets 📊

You'll work with the following datasets across the sessions:

Dataset	Task
Higgs Boson Dataset	Binary classification with high-energy physics data
Heart & Lung Sounds (HLS-CMDS)	Anomaly detection using autoencoders
NEU Surface Defects Dataset	CNN-based image classification
ISIC Skin Lesion Dataset	Medical image segmentation using U-Net and transfer learning

---

Repository Structure 📁

UoM_fse_dl_workshop/
├── SE##_CA_*.ipynb      # Code-along notebooks for live exercises
├── solutions/           # Completed notebooks with full implementations
├── figs/                # Figures and diagrams
└── utils/               # Helper tools used throughout the workshop
    ├── plotting/
    ├── data/
    ├── ml/
    └── solutions.json

---

Prerequisites 🔧

• Basic Python programming skills
• Familiarity with Jupyter Notebooks, NumPy, and Pandas
• Understanding of core ML concepts (e.g., loss functions, model evaluation)
• No PyTorch experience required!
• (Optional) Background in linear algebra and calculus is beneficial

---

Additional Resources 📚

PyTorch & Models

• PyTorch Docs
• torchvision Models
• PyTorch Hub
• Hugging Face Models

Visual Tools

• CNN Explainer
• Distill Feature Visualization
• TensorBoard for PyTorch

Courses & Books

• Deep Learning with PyTorch (60-min Blitz)
• Fast.ai Course
• Dive into Deep Learning