Yuandong Tian web metrics

Research Scientist and Manager
Facebook Artificial Intelligence Research
Email: yuandong [at] fb [dot] com


Brief Bio

I am a Research Scientist and Manager in Facebook AI Research (FAIR). I received Ph.D in the Robotics Institute, Carnegie Mellon University. Recipient of ICCV 2013 Marr Prize Honorable Mentions.


Google Scholar, CV

Reseach Directions

Reinforcement Learning and Search, Representation Learning

Talks

Apr. 21, 2021 Invited Guest Lecture in UPenn (Thanks Jing Li for the invitation). Slides here.

Apr. 21, 2021 Invited talk in VALSE Webinar about understanding self-supervised learning. Slides here.

Apr. 12, 2021 Invited Talk in UCL DARK Lab. Slides here.

Dec. 12, 2020 Contributed talk in NeurIPS 2020 workshop of Self-supervised Learning, Theory and Practice.

Dec. 12, 2020 Invited talk in NeurIPS 2020 workshop of Learning meets Combinatorial Algorithms.

Nov. 30, 2020 Invited Talk (Slides) at Workshop of Reinforcement Learning from Batch Data and Simulation in Simons Institute of UC Berkeley

Oct. 20, 2020 Invited Guest Lecture in University of Wisconsin Madison (Class syllabus: link)

Oct. 14, 2020 Distinguished Guest Lecture in IIIS, Tsinghua University.

Jun. 6, 2020 Invited guest lecture in UCLA (online)

Nov. 7, 2019 Invited Talk in IAS "Workshop on New Directions in Reinforcement Learning and Control" in Princeton University.

Nov. 6, 2019 Invited Talk in NEC Laboratories Princeton.

Oct. 27, 2019 Invited Talk in AI Sys Workshop in SOSP'19.

Jun. 2019 Long oral talk about OpenGo in ICML 2019.

Jan. 2019 Talks in Deep Learning Summit, AAAI 2019 Workshops (Reproducible AI and Game and Environments in Artificial Intelligence)

Jan.-Jun. 2018. Multiple talks in Stanford, AI NextCon, etc. link

Dec. 20, 2017. Keynote at Future Leaders of AI Retreat (FLAIR), Shanghai. link

Dec. 6, 2017. Oral talk about ELF platform, NIPS 2017, Long Beach. link

Nov. 5, 2017. DRL and Game Tutorial in AI Frontier, Santa Clara. link.

Oct. 27, 2017. DRL and Game Tutorial in Montain View, ACMMM 2017 Slides.

Aug. 10, 2017. Presentation in Video Games and Machine Learning (VGML) Workshop, ICML 2017. Slides. The same talk is also presented in University of Sydney on Aug. 11, hosted by Dong Xu.

Aug. 8, 2017. "An Analytical Formula of Population Gradient for two-layered ReLU network and its Applications in Convergence and Critical Point Analysis", ICML 2017. Slides

Jun. 30, 2017 - Jul. 9, 2017. "AI In Games: Achievements and Challenges", 5 talks in China (Beijing, Shanghai, Shenzhen). Slides.
  • Jun. 30, 2017, Chinese Academia of Science, Institute of Automation (CASIA)
  • Jun. 30, 2017, Tsinghua University, State Key Laboratory of Intelligent Technology and Systems
  • Jul. 4, 2017. ShanghaiTech Symposium on Information Science and Technology (SSIST) 2017. link
  • Jul. 6, 2017. Brain-AI workshop. link
  • Jul. 9, 2017. CCF-GAIR 2017 link

Projects

Reinforcement Learning and Search


BeBold: Exploration Beyond the Boundary of Explored Regions [arXiv][3 min video]
Tianjun Zhang, Huazhe Xu, Xiaolong Wang, Yi Wu, Kurt Keutzer, Joseph E. Gonzalez, Yuandong Tian
arXiv 2020

Multi-Agent Collaboration via Reward Attribution Decomposition [arXiv][Code][3 min video][Website]
Tianjun Zhang, Huazhe Xu, Xiaolong Wang, Yi Wu, Kurt Keutzer, Joseph E. Gonzalez, Yuandong Tian
arXiv 2020

Joint Policy Search for Multi-agent Collaboration with Imperfect Information [arXiv][Code][3 min video]
Yuandong Tian, Qucheng Gong, Tina Jiang
NeurIPS 2020

Learning Search Space Partition for Black-box Optimization using Monte Carlo Tree Search [arXiv][Code]
Linnan Wang, Rodrigo Fonseca, Yuandong Tian
NeurIPS 2020

Sample-Efficient Neural Architecture Search by Learning Action Space [arXiv][Code]

IEEE Transactions on Pattern Analysis and Machine Intelligence (T-PAMI), 2021

Linnan Wang, Saining Xie, Teng Li, Rodrigo Fonseca, Yuandong Tian

ELF OpenGo: An Analysis and Open Reimplementation of AlphaZero [arXiv][Code][Pre-trained models and game records][Blogpost][Talk]

A open source reimplementation of DeepMind's zero-knowledge training and its application to the game of Go. Trained on 2000 GPUs for 9 days. With a single GPU and 50 seconds per move, the model won 20-0 versus 4 top 30 professional players, given human unlimited thinking time. It also won 980-18 versus LeelaZero (version Apr. 25).

ICML 2019 (Long Oral)

Yuandong Tian, Jerry Ma*, Qucheng Gong*, Shubho Sengupta*, Zhuoyuan Chen, James Pinkerton, Larry Zitnick

Algorithmic Framework for Model-based Deep Reinforcement Learning with Theoretical Guarantees [Link][Code]
Yuping Luo, Huazhe Xu, Yuanzhi Li, Yuandong Tian, Trevor Darrell, Tengyu Ma
ICLR 2019

M^3RL: Mind-aware Multi-agent Management Reinforcement Learning [Link][Code]
Tianmin Shu, Yuandong Tian
International Conference on Learning Representations (ICLR), 2019

Building Generalizable Agents with a Realistic and Rich 3D Environment [Arxiv Link][Code]
Yi Wu, Yuxin Wu, Georgia Gkioxari, Yuandong Tian
Workshop in International Conference on Learning Representations (ICLR), 2018

ELF: An Extensive, Lightweight and Flexible Research Platform for Real-time Strategy Games [Link][Repo][Demo]
Yuandong Tian, Qucheng Gong, Wenling Shang, Yuxin Wu, Larry Zitnick
Advances in Neural Information Processing Systems (NIPS), 2017
Oral presentation[link]

Training Agent for First-Person Shooter Game with Actor-Critic Curriculum Learning [Link]
Yuxin Wu, Yuandong Tian
International Conference on Learning Representations (ICLR), 2017

Better Computer Go Player with Neural Network and Long-term Prediction [arXiv Page] [poster] [Code] [Pre-trained Models]
Yuandong Tian, Yan Zhu
International Conference on Learning Representations (ICLR), 2016

Theoretical Understanding of Deep Networks


Understanding Self-supervised Learning Dynamics without Contrastive Pairs [arXiv][3min video]
Yuandong Tian, Xinlei Chen, Surya Ganguli

Understanding Self-supervised Learning with Dual Deep Networks [arXiv][3min video]
Yuandong Tian, Lantao Yu, Xinlei Chen, Surya Ganguli


Understanding Robustness in Teacher-Student Setting: A New Perspective
Zhuolin Yang*, Zhaoxi Chen, Tiffany Cai, Xinyun Chen, Bo Li, Yuandong Tian* (*=Equal Contribution)
AIStats 2021


Student Specialization in Deep ReLU Networks With Finite Width and Input Dimension [arXiv][Talk][Code]
Yuandong Tian
ICML 2020


Luck Matters: Understanding Training Dynamics of Deep ReLU Networks [arXiv][Workshop Poster]

Yuandong Tian, Tina Jiang, Qucheng Gong, Ari Morcos
ICML 2019 workshop on Understanding and Improving Generalization in Deep Learning

Playing the lottery with rewards and multiple languages: lottery tickets in RL and NLP [arXiv]

Haonan Yu, Sergey Edunov, Yuandong Tian, Ari S. Morcos
ICLR 2020

One ticket to win them all: generalizing lottery ticket initializations across datasets and optimizers [arXiv]

Ari S. Morcos, Haonan Yu, Michela Paganini, Yuandong Tian
NeurIPS 2019

A Theoretical Framework for Deep Locally Connected ReLU Network [ArXiv link][Poster]
Yuandong Tian
Arxiv preprint, 2018

Gradient Descent Learns One-hidden-layer CNN: Don't be Afraid of Spurious Local Minima [ArXiv link]
Simon S. Du, Jason D. Lee, Yuandong Tian, Barnabas Poczos, Aarti Singh
International Conference on Machine Learning (ICML), 2018

When is a Convolutional Filter Easy To Learn? [ArXiv link][OpenReview]
Simon S. Du, Jason D. Lee, Yuandong Tian
International Conference on Learning Representations (ICLR), 2018

An Analytical Formula of Population Gradient for two-layered ReLU network and its Applications in Convergence and Critical Point Analysis [Link][ICLR Workshop version][Code]
Yuandong Tian
International Conference on Machine Learning (ICML), 2017

Vision and Question Answering

Semantic Amodal Segmentation [arXiv Page]
Yan Zhu, Yuandong Tian, Dimitris Mexatas, Piotr Dollár
Computer Vision and Pattern Recognition (CVPR), 2017
Single Image 3D Interpreter Network [arXiv Page]
Jiajun Wu, Tianfan Xue, Joseph J. Lim, Yuandong Tian, Joshua B. Tenenbaum, Antonio Torralba, William T. Freeman
European Conference on Computer Vision (ECCV), 2016
Oral presentation
Simple Baseline for Visual Question Answering [arXiv Page]
Bolei Zhou, Yuandong Tian, Sainbayar Sukhbaatar, Arthur Szlam, Rob Fergus
ArXiv preprint, 2016

PhD thesis

Theory and Practice of Globally Optimal Deformation Estimation[PDF]
Yuandong Tian, PhD thesis, CMU-RI-TR-13-25

Research Projects in CMU

Theory and Practice of Hierarchical Data-driven Descent for Optimal Deformation Estimation[PDF]
Yuandong Tian, Srinivasa G. Narasimhan,
International Journal of Computer Vision (IJCV), 2015
Hierarchical Data-Driven Descent for Efficient Optimal Deformation Estimation[PDF][Detailed Proofs]
Yuandong Tian, Srinivasa G. Narasimhan,
ICCV 2013 Marr Prize Honorable Mentions
Exploring the Spatial Hierarchy of Mixture Models for Human Pose Estimation [PDF] [Evaluation Code][Project Page]
Yuandong Tian, C. Lawrence Zitnick, Srinivasa G. Narasimhan,
Accepted as Poster in ECCV 2012
Learning from Crowds in the Presence of Schools of Thought [PDF][Dataset][Code][Presentation]
Yuandong Tian, Jun Zhu,
ACM SIGKDD 2012
Depth from Optical Turbulence [Project Page]
Yuandong Tian, Srinivasa G. Narasimhan, Alan J. Vannevel
Accepted as Poster in CVPR 2012
Globally Optimal Estimation of Nonrigid Image Distortion[PDF]
Yuandong Tian, Srinivasa G. Narasimhan,
International Journal of Computer Vision (IJCV), As an extended version of [Tian and Narasimhan, CVPR 2010].
Rectification and 3D reconstruction of Curved Document Images [Project Page]
Yuandong Tian, Srinivasa G. Narasimhan,
Oral presentation in CVPR 2011

Local Isomorphism to Solve the Pre-image Problem in Kernel Methods
Dong Huang, Yuandong Tian, Fernando De la Torre
Accepted as Poster in CVPR 2011
A Globally Optimal Data-Driven Approach for Image Distortion Estimation [Project Page]
Yuandong Tian, Srinivasa G. Narasimhan,
Oral presentation in CVPR 2010

Seeing through water: Image restoration using model-based tracking [Project Page]
Yuandong Tian, Srinivasa G. Narasimhan,
Accepted as Poster in ICCV 2009
(De) Focusing on Global Light Transport for Active Scene Recovery
Mohit Gupta, Yuandong Tian, Srinivasa G. Narasimhan, Li Zhang
Oral presentation in CVPR 2009
[PDF]



Relationship between projector defocus and global illumination for statistically-modeled scenes. [Project Page]
Yuandong Tian, Mohit Gupta, Srinivasa G. Narasimhan, Li Zhang
Technical Report CMU-RI-TR-09-10, Carnegie Mellon University, March 2009.

Previous Research Projects

Easytoon: an easy and quick tool to personalize a cartoon storyboard using family photo album
Shifeng Chen, Yuandong Tian, Fang Wen, Ying-Qing Xu, Xiaoou Tang
ACMMM 2008 [PDF]
A Face Annotation Framework with Partial Clustering and Interactive Labeling
Yuandong Tian, Wei Liu, Rong Xiao, Fang Wen, Xiaoou Tang
Accepted as Poster in CVPR 2007 [PDF]

Notes

Short Notes on Graphcut and MRF
Yuandong Tian, 2009
[PDF]
A short introduction to Riemann Geometry
Yuandong Tian, 2009
[PDF]
Some notes in Pattern Recognition and Machine Learning
Yuandong Tian, 2008
[PDF]

Personal

Poems and Novels

Homebrew programming on Sony PSP (Playstation Portable)