Hands-on Introduction to PyTorch for Machine Learning

Overview
This 4 hour workshop will introduce students to using PyTorch for Machine Learning. The class will be taught by Graham Ganssle of Expero.

The course will use the example of an introductory customer journey. The problem presented will be as such:
If Sally purchases a mattress,two trash cans, and a couch, then she may fit into a “new homeowner” customer classification. Perhaps a new home owner also needs a dish set! But more than just cross-selling needed home goods, can we start to establish a holistic view of Sally? A view that will guide us on how to be a better company that serves our customers better? The technique covered in this four hour mini-course is the first building block for
creating a comprehensive view of a company’s customers. This stepping stone is fundamental in predicting customers’ next actions.

In this course we will learn the basics of PyTorch, including:
● Tensors
● Variables
● Automatic gradient calculation
● Back - Propagation
● Building and training simple neural networks with hidden layers.

Students are advised to bring a laptop and code along, although it should be noted that PyTorch is currently only supported in Linux and Mac OS (not Windows). The course assumes no previous knowledge of PyTorch, and though some understanding of neural networks is encouraged, the instructor will diagram the network architecture and discuss its use on the whiteboard.

About the Instructor

Graham Ganssle loves data. His favorite part of work at Expero is daydreaming up innovative solutions to quantifiable problems and planning an implementation strategy. Building intelligent systems is his passion whether it’s automated derivatives trading bots, adaptive image processing algorithms, or autonomous musical composers. Whether deep learning is the optimal solution or not, helping customers succeed through solving their analytics problems is where Graham finds the most satisfaction.

Graham Ganssle’s physics Ph.D. focused on digital signal processing, specifically on a (then) new optimization method which used naturally coupled wavefields to stabilize convergence. He also holds a masters degree in applied physics and a professional geoscientist license. Graham worked in the oil and gas vertical for ten years, performing data science and quantitative geophysics for clients around the world. He has numerous publications on a variety of scientific topics and has been awarded both scientific and business achievement awards.

Off the clock, Graham’s an inept aspiring rock climber and a triathlete. He’s constantly imploring his bride (and, more successfully, his puppy) to get muddy with him on the trail. Most Saturday mornings you can find Graham clacking away at his keyboard on his newest experiment or craziest inspiration.