CNNS

Intro to Convolutional Neural Networks in PyTorch

A Comprehensive Guide to Understanding and Implementing CNNs in PyTorch for Computer Vision

Table of Contents

1. About The Workshop
2. Getting Started
   • Dependencies
   • Installation
   • Usage
3. Workshop Content
   • Convolutional Layers
   • Pooling Layers
   • Simple CNN Model
   • Training a CNN
   • MNIST Dataset
4. Roadmap
5. License
6. Authors
7. Acknowledgements

About The Workshop

This workshop introduces participants to Convolutional Neural Networks (CNNs), the backbone of many computer vision models in deep learning. Using PyTorch, we'll explore the architecture and functionality of CNNs, including convolutional layers, pooling layers, and training on the MNIST dataset.

Getting Started

Participants will use Jupyter Notebooks to interact with CNNs in PyTorch and apply the learned concepts through hands-on exercises.

Dependencies

• A working computer

Installation (only if running locally, not needed if using the Google Collab)

1. Clone the workshop repository:

   git clone https://github.com/matheusmaldaner/WorkshopArchive.git

2. Install the required Python libraries if running locally:

   pip install torch torchvision matplotlib jupyter

Usage

Open the Jupyter Notebook provided in the repository or use Google Colab to participate in the workshop. Each section contains exercises and examples that guide you through CNN functionalities in PyTorch.

Workshop Content

Convolutional Layers

Explore the creation and function of convolutional layers, the backbone of CNNs. Understand how these layers use filters to detect features in images.

Pooling Layers

Learn about pooling layers, which reduce the dimensions of feature maps, decreasing computational load and minimizing overfitting.

Simple CNN Model

Build a simple CNN combining convolutional, pooling, and fully connected layers to classify images.

Training a CNN

Learn how to train a CNN by defining loss functions, optimizers, and iterating through data loops using PyTorch.

MNIST Dataset

Apply the learned concepts by training a CNN on the MNIST dataset, the "Hello World" of machine learning for handwritten digit classification.

Roadmap

1. Introduction to CNNs and their role in computer vision.
2. Detailed exploration of convolutional and pooling layers.
3. Construction and training of a simple CNN model.
4. Application on the MNIST dataset with accuracy evaluation.

License

Distributed under the MIT License. See LICENSE for more information.

Authors

Jim Su - @yimothysu

Matheus Kunzler Maldaner - @matheusmaldaner

Acknowledgements

• Data Science and Informatics Institute for the room

Thank You