Deep Learning Development with PyTorch+Jupyter using Heterogenous hardware

Joseph Spisak

Audience level:: Intermediate

Brief Summary

PyTorch + Jupyter provides an interactive and engaging way to do deep learning development. Additionally, heterogenous hardware, such as Cloud TPUs, continue to permeate the ecosystem and become more easily accessible to the developer community. In this talk, Joe Spisak - PyTorch Product Lead, will walk you through the process of building machine learning models with PyTorch in Jupyter.

Outline

Intro to PyTorch - what it is, who uses it, why it has such an amazing community and how we use Jupyter!
Developing a model with Jupyter on Colab - data handling, model development and training and visualizing loss curves using TensorBoard, model evaluation.
Interpreting the model with Captum and understanding what is attributing to the predictions
Benchmarking different hardware backends using CPUs, GPUs and Cloud TPUs