Welcome!

This is the homepage for Garrett Thomas. I recently graduated with a degree in Computer Science and Mathematics from UC Berkeley and will begin my PhD in Computer Science at Stanford in Fall 2018. My academic interests include machine learning (particularly deep learning and reinforcement learning), the math behind it, and education.


Research

Robot Learning Lab

I was part of Pieter Abbeel's group, working with postdoc Aviv Tamar on deep reinforcement learning with applications to robotics. In particular, we investigated questions regarding policy representation, planning, and generalization.

Oski Lab

Freshman year I worked with Cyrus Dioun in Oski Lab. I wrote Python scripts for web scraping and data analysis.

Papers

Teaching

Fall 2016, Spring 2017, Fall 2017

I was an Undergraduate Student Instructor for CS 189, Introduction to Machine Learning. I have written (or rather, am writing) Mathematics for Machine Learning to help fill gaps in students' mathematical background, particularly in linear algebra, optimization, and probability.

Deep RL Bootcamp

I was thrilled to have the opportunity to work as a Teaching Assistant for the Deep RL Bootcamp. This two-day introduction to deep reinforcement learning includes a mixture of theory and practice, aiming to expose attendees to a range of topics from the basics to current research frontiers.

Spring 2016

I was an Undergraduate Student Instructor for Data 8, Foundations of Data Science. This exciting new class teaches modern statistical inference and data analysis with a strong focus on computing.

Fall 2015

I was a Tutor for CS 61A, Structure and Interpretation of Computer Programs. This very large and excellent class provides an introduction to computer programming, emphasizing abstractions and giving an overview of several programming paradigms.


Projects

gtml

I am developing this library to support my own machine learning research. It features tight integration with the powerful TensorFlow library for constructing and optimizing neural networks. These can be trained using supervised or (soon) reinforcement learning methods.

illume

A number of years ago (in high school) I developed the puzzle game illume. It is completely free, has been downloaded over 22,000 times, and has received many positive reviews, so it might be worth checking out if you have an iOS device. If you hate it, I'd love to hear feedback about that as well.


Contact

Feel free to reach out to me via email: gwthomas@stanford.edu.