Cornell Virtual Workshop: Building Scalable CNN Models

Roadmap: Building Scalable CNN Models

Topics

1. Image Classification and CNN 2. Building a CNN Classifier with PyTorch: Part 1 3. Building a CNN Classifier with PyTorch: Part 2 4. Introduction to DDP with PyTorch 5. Single Node Multi-GPU Training with torchrun 6. Multi-node Distributed Training with PyTorch Lab Setup

This course is designed to teach you how to build scalable CNN classifiers for building damage identification and interpret these models.

These course materials are made available through Chishiki-AI, a transformative project at the forefront of integrating Artificial Intelligence (AI) with Civil and Environmental Engineering (CEE). Chishiki.AI, meaning ‘knowledge through AI,’ is a pioneering initiative funded by the National Science Foundation and led by a team of experts at the University of Texas at Austin in collaboration with Cornell University.

Chishiki-AI is funded by the National Science Foundation, award #2321040.

Objectives

After you complete this workshop, you should be able to:

Explain the fundamentals of building image classification models as well as techniques to train models with data parallelism
Build, train, and evaluate a CNN image classification model on a single GPU with PyTorch
Migrate a single GPU PyTorch training script to run on multiple GPUs and multiple nodes in a cluster using distributed data parallel

Prerequisites

The prerequisites for the topics in CNN are:

Experience with programming in Python.

Requirements

To perform the exercises outlined in this module you will need an allocation on TACC resources with access to the TACC Analysis Portal.

If you don't have an allocation: Register with the Chishiki-AI project to be added to the TACC account.