I am a (final year) CS PhD student studying artificial intelligence at Stanford University, advised by Chelsea Finn. Email: ayz@cs.stanford.edu. Also, see my CV (PDF link).
Research: I am interested in scalable techniques that improve the training of deep neural networks. This includes (1) hyperparameter optimization and learned optimizers, and (2) designing deep architectures that guarantee useful symmetries.
Summary: Most recently: student researcher at Google DeepMind designing learned optimizers. Previously: RL research intern at Meta AI, AI resident with Google Brain robotics, undergrad robotics research with Anca Dragan at UC Berkeley.
Some papers (all)
- Universal Neural Functionals
Allan Zhou, Chelsea Finn, James Harrison.
Preprint. [arXiv] [Code] - Permutation Equivariant Neural Functionals
Allan Zhou, Kaien Yang, Kaylee Burns, Adriano Cardace, Yiding Jiang, Samuel Sokota, J. Zico Kolter, Chelsea Finn.
NeurIPS 2023. [arXiv] [Code] - NeRF in the Palm of Your Hand: Corrective Augmentation for Robotics via Novel-View Synthesis
Allan Zhou*, Moo Jin Kim*, Lirui Wang, Pete Florence, Chelsea Finn.
CVPR 2023. [arXiv][Videos] - Fleet Policy Learning via Weight Merging and An Application to Robotic Tool-Use
Lirui Wang, Kaiqing Zhang, Allan Zhou, Max Simchowitz, Russ Tedrake.
ICLR, 2024. [arXiv]
Software
- UNF: A JAX library for building equivariant neural networks that process any weight-space features, or any collection of permutable tensors.
- NFN: A PyTorch library for building equivariant neural networks that process MLP and CNN weight-space features.
- midGPT: A simple JAX + Equinox implementation of a modern large language model (LLM), for quick experimentation and research. Easily scale to billions of parameters on cloud TPU slices.
Teaching
- TA, Stanford CS230 (Deep Learning). Spring 2022.
- TA, Stanford CS224N (Natural Language Processing w/ Deep Learning). Winter 2021-2022.
- TA, UC Berkeley CS 188 (Intro to Artificial Intelligence). Spring 2018.