ChainerRL is a deep reinforcement learning library built upon the Chainer deep learning library. Part of my work at Preferred Networks is devoted to developing algorithms and infrastructure for ChainerRL.
A high-quality implementation of Human-level control through deep reinforcement learning. The is a deterministic implementation, aimed at reproducibility. This was built with PyTorch.
This project replicates Google Deepmind's Deep Q-Network (DQN) agent as in their Nature Paper. This project is written in python using the Arcade Learning Environment and the Tensorflow library. Also a special thanks to Josh Kelle for his help and insights and for teaching me new things about python and tensorflow.
Inverse Reinforcement Learning aims to address the difficult problem in reinforcement learning of defining a reward function. Essentially, a demonstrator provides multiple demonstrations of the task to the learning agent. The learning agent uses this information to learn a reward function for the task, after which standard reinforcement learning algorithms can be applied.
My work
My project in inverse reinforcement learning stems from a graduate Robot Learning class I took with Professor Scott Niekum at the University of Texas at Austin. In this project, we implement a core inverse reinforcement learning algorithm, Maximum Entropy Inverse Reinforcement Learning. Additionally, we implement a recent paper Inverse Reinforcement Learning from Failure. We expand on both these ideas by developing and implementing two new algorithms Inverse Reinforcement Learning from Ranking and Inverse Reinforcement Learning from Failure and Ranking. This work was done in Python.
Important Links
Please reach out to me if you would like to see the code.
In this work we teach an agent to learn to replicate demonstrated trajectories in 2 dimensions. Furthermore, it can learn to generalize these trajectories to new goals. This was done in Java.
Use this area of the page to describe your project. The icon above is part of a free icon set by Flat Icons. On their website, you can download their free set with 16 icons, or you can purchase the entire set with 146 icons for only $12!
