Josh Tobin is a Research Scientist at OpenAI and a PhD student in Computer Science at UC Berkeley working with Professor Pieter Abbeel. Josh's research focuses on applying deep learning to problems in robotic perception and control, with a particular concentration on deep reinforcement learning generative models, and synthetic data. Prior to Berkeley and OpenAI, Josh was a consultant at McKinsey & Co. in New York. Josh has a BA in Mathematics from Columbia University.
Troubleshooting Deep Neural Networks
When I started in deep learning, I felt frustrated that I was spending most of my time debugging instead of the "fun" stuff. (Later, I discovered that debugging never goes away, and the best practitioners still spend most of their time on it.)
As I learned more and began helping others train models, I realized that much of my advice consisted of walking people through a mental decision tree for how to improve their model's performance.
This guide is an attempt to codify that decision tree.