Introduction to PyTorch: Completely uninformed image modeling

PyTorch (http://pytorch.org/) is a GPU matrix library for Python.

Based on the ideas (and C code) for Torch, a lua environment for scientific computation, it combines speed of execution with an elegant programming framework for computation on the GPU.

Pytorch has specialised modules for automatic differentiation,neural networks and transforms

• In this talk we will:

• Introduce the core library and concepts behind PyTorch

• Cover simple linear models with PyTorch

• Look at how to import data into PyTorch format

• Fit some (relatively) simple neural networks on image data

• Fit some LSTM's to text data

• Examine the performance relative to traditional approaches

• We will use samples from the Kaggle rentals (https://www.kaggle.com/c/two-sigma-connect-rental-listing-inquiries) competition

• Specifically, a sample of images obtained during the course of my last talk (https://www.meetup.com/DublinR/events/239190623/)

Speaker: Richie Morrisroe