Tongzhou Wang

Tongzhou Wang
王童舟

Structures + ML + AI

I am… a final-year PhD student at MIT with Antonio Torralba and Phillip Isola. I work on machine learning, representation learning, and reinforcement learning.

Research. I am interested in inductive structures in machine learning and artificial intelligence, with a focus on useful structures for better perception and decision-making and learning such structures with neural nets (e.g., invariances to imperceptible changes, distances w.r.t. decision-making capabilities, task-specific factorizations of signal and noise).

Most of my works are related to these specific topics:

  • Learning fundamental structures for better AI systems, with theoretical guarantees and empirical benefits.

  • Analyzing and discovering useful structures, e.g., what structure an algorithm learns, what makes for a good training set.

Outside research, I spent my time on developing 1st offering of MIT’s Deep Learning course, GitHub User's stars open-source ML projects, organizing a NeurIPS workshop on Goal-Conditional RL, mentoring SGI students (blog), pro bono office hours (book me!), and with 😸😼.

On 2023-2024 faculty job market: YES ✅

Email: tongzhou _AT_ mit _DOT_ edu

Selected Publications (full list)

Learned structure for better agents Analyze and discover useful structures

Optimal Goal-Reaching Reinforcement Learning via Quasimetric Learning
[ICML 2023][Project Page] [arXiv] [Code]
Tongzhou Wang, Antonio Torralba, Phillip Isola, Amy Zhang
paper thumbnail Quasimetric Geometry +
Novel Objective
(Push apart s_start & goal
while keeping local dists.)
 = Optimal Value V∗
AND
High-Performing
Goal-Reaching Agents

Improved Representation of Asymmetrical Distances with Interval Quasimetric Embeddings
[NeurIPS 2022 NeurReps Workshop] [Project Page] [arXiv] [PyTorch Package for Quasimetric Learning]
Tongzhou Wang, Phillip Isola
computing-iqe

Denoised MDPs: Learning World Models Better Than The World Itself
[ICML 2022] [Project Page] [arXiv] [code]
Tongzhou Wang, Simon S. Du, Antonio Torralba, Phillip Isola, Amy Zhang, Yuandong Tian

On the Learning and Learnability of Quasimetrics
[ICLR 2022] [Project Page] [arXiv] [OpenReview] [code]
Tongzhou Wang, Phillip Isola
quasimetric-function-spaces
Learning to See by Looking at Noise
[NeurIPS 2021] [Project Page] [arXiv] [code & datasets]
Manel Baradad*, Jonas Wulff*, Tongzhou Wang, Phillip Isola, Antonio Torralba
learning-to-see-by-looking-at-noises

Understanding Contrastive Representation Learning through Alignment and Uniformity on the Hypersphere
[ICML 2020] [Project Page] [arXiv] [code]
Tongzhou Wang, Phillip Isola
hypersphere_stl10_scatter_linear_output 
# bsz : batch size (number of positive pairs)
# d   : latent dim
# x   : Tensor, shape=[bsz, d]
#       latents for one side of positive pairs
# y   : Tensor, shape=[bsz, d]
#       latents for the other side of positive pairs
def align_loss(x, y, alpha=2):
  return (x - y).norm(p=2, dim=1).pow(alpha).mean()

def uniform_loss(x, t=2):
  return torch.pdist(x, p=2).pow(2).mul(-t).exp().mean().log()
PyTorch implementation of the alignment and uniformity losses

Dataset Distillation
[Project Page] [arXiv] [code] [DD Papers]
Tongzhou Wang, Jun-Yan Zhu, Antonio Torralba, Alexei A. Efros
dataset_distillation_fixed_mnist