John Luoyu Zhou

Hello! I'm a Computer Science Ph.D. student at UCLA, advised by Jonathan Kao. I received my B.S., also in Computer Science, from Columbia University, where I was advised by Liam Paninski in the Center for Theoretical Neuroscience.

CV  /  Twitter  /  GitHub  /  Google Scholar

Current Interests

I am focused on developing reinforcement learning algorithms that are able to reason through complex, long-horizon tasks by recursively decomposing them into simpler subproblems. To this end, I work on hierarchical methods that operate over (sub)goal-conditioned policies, action chunks, and options.

I spent my undergraduate studies and the first part of my Ph.D. studying neuroscience, investigating how the brain allocates control between model-free and model-based decision-making strategies and developing neural decoding applications.

Publications

Flattening Hierarchies with Policy Bootstrapping

We present Subgoal Advantage-Weighted Policy Bootstrapping (SAW), an offline goal-conditioned RL algorithm that achieves state-of-the-art performance on complex, long-horizon tasks without needing hierarchical policies or generative subgoal models.

Reciprocal Reward Influence Encourages Cooperation From Self-Interested Agents

We propose a form of intrinsic motivation to reciprocate the influence of other agents’ actions on one’s own return, and show that this is sufficient to encourage mutual cooperation from other, self-interested agents in sequential social dilemmas.

Previous Research

Neuroscience Cloud Analysis As a Service

An open-source, drag-and-drop platform that uses cloud resources to run modern neuroscience data analyses.

Partitioning variability in animal behavioral videos using semi-supervised variational autoencoders

A semi-supervised framework that combines the output of supervised pose estimation algorithms with unsupervised dimensionality reduction methods to produce interpretable, low-dimensional representations of behavioral videos.

Writing

A Full Dive

An article that takes a dive into the current state of brain-computer interfaces and open challenges in neural decoding and encoding. Inspired by my research as well as Sword Art Online, one of my favorite childhood anime series.

Template