I’d give a talk on building custom deep recommendation engines. It’s hands-on tutorial that very briefly touches on the math of matrix-factorization approaches that power recommendation engines at Stitch Fix, Amazon, Pinterest and Netflix. We’ll get our hands dirty with PyTorch and show how to make your own customized models—a great example of how deep learning enables not just powerful models, but ones that are easily tailored to individual challenges. It’ll be about 15% math, 85% PyTorch code.

I’ll start with introducing the concepts, building a simple matrix factorization model in PyTorch in <30 lines of code. I’ll show how to train the model, criticize it and customize it to fit the characteristics of your specific problem. By the end, someone will start with a simple factorization model, and optionally upgrade it to a factorization machine (which easily responds to new feature interactions), or to a variational factorization machine (which sits at the heart of Bayesian Deep Learning) or a sparse matrix factorization machine (that is best for interpreting & explaining models to human). Either way, folks will have a state of the art deep learning recommendation model and will be in a good place to continue tinkering with PyTorch.

Time and location will be posted when available.

Return to: Hands-on Tutorials