Teaching a neural network to use a calculator
This article explores a seq2seq architecture for solving simple probability problems in Deepmind’s Mathematics Dataset. A transformer is used to map questions to intermediate steps, while an external symbolic calculator evaluates intermediate expressions. This approach emulates how a student might solve math problems, by setting up intermediate equations, using a calculator to solve them, and using those results to construct further equations.
Neural Style Transfer with Adversarially Robust Classifiers
I show that adversarial robustness makes neural style transfer work on a non-VGG architecture.
Teaching agents to paint inside their own dreams
In this post, I talk about using World Models to train agents to paint with real painting software. This includes my thought process, approach, failures, and some future work in this direction I’m excited about.
Porting arbitrary style transfer to the browser
Recently, Magenta, Google’s “open source research project exploring the role of machine learning as a tool in the creative process” gave me the opportunity to write a blog post on their platform about my recent work porting arbitrary style transfer to the browser. Check it out here!