Lecture & Practical Session on Advanced CNNs

Date:

In April 2025, together with my master students Claudia Castrillón-Álvarez and Laura Rodrigo-Muñoz, I delivered a session titled “Lecture & Practical Session on Advanced CNNs” at the Ágora Journal Club organized by the Signal Theory and Communications Department at Universidad Carlos III de Madrid.

The session included:

  • 📊 30-minute theory talk on:
    • CNN building blocks (convolutions, pooling, activations, batch norm)
    • Architectures: ResNet, U-Net, YOLO
    • Transfer learning and fine-tuning
    • Evaluation metrics (accuracy, F1, mAP, Dice, IoU)
  • 🧪 Hands-on Jupyter Notebook covering:
    • Classification with custom CNNs and pretrained ResNet
    • Object detection with YOLOv8
    • Segmentation with U-Net and MONAI UNet
    • Tailored loss functions and evaluation strategies

🧠 Datasets Used:

  • Brain Cancer - MRI Dataset (Kaggle)
  • Brain Tumor Detection Dataset (Ultralytics)
  • Pancreatic Stem Cells Dataset (Cell Tracking Challenge)

📁 Materials: Zenodo (10.5281/zenodo.15276042)
💻 GitHub Repository: Ágora Sessions (update with real URL if needed)

This session aimed to give students an integrated perspective on how CNNs are used across classification, detectio