I’m an assistant professor in Westlake University (西湖大学) (AI direction in Department of Engineering), PI of AI for Scientific Simulation and Discovery Lab（人工智能与科学仿真发现实验室）. I did my postdoc at Stanford Computer Science (2020 to mid-2023), working with professor Jure Leskovec, got my Ph.D. in MIT in 2019, and B.S. in Peking University in 2012. I’m now actively recruiting postdocs, PhD students and interns. Toward my long-term vision of accelerating scientific simulation, design and discovery with machine learning (AI + Science), my research interests are (my CV):

(1) AI for scientific simulations & design: develop machine learning (ML) techniques (GNNs, diffusion, etc.) to accelerate scientific simulations and design. Application in fluid, plasma physics, materials, life science, etc. with large-scale (up to millions of nodes) and multi-scale characteristics.
(2) AI for scientific discovery: discover scientific theories (compact governing laws) and concepts (with relational structures) from noisy observational data, with neuro-symbolic AI. Application in life science, physics and other sciences.
(3) Representation learning: develop novel representation learning algorithms to learn structured representations from data, based on information theory, GNNs and diffusion models, inspired by important questions in AI + Science.

I obtained my PhD from MIT (thesis: Intelligence, physics and information – the tradeoff between accuracy and simplicity in machine learning), where I worked on AI for Physics and Physics for AI, with Prof. Isaac Chuang and Prof. Max Tegmark. I also joyfully interned at Google AI and NVIDIA Research.

Group Homepage: AI for Scientific Simulation and Discovery Lab（人工智能与科学仿真发现实验室）
Contact: tailin@cs.stanford.edu
Google Scholar | Twitter | GitHub
Group Github | University Homepage

News:

Publications:

Compositional Generative Inverse Design
Tailin Wu*, Takashi Maruyama*, Long Wei*, Tao Zhang*, Yilun Du*, Gianluca Iaccarino, Jure Leskovec
ICLR 2024 Spotlight
[paper][arXiv][code][slides]
TL;DR: [ML for design] We introduce a method that uses compositional generative models to design boundaries and initial states significantly more complex than the ones seen in training for physical simulations

BENO: Boundary-embedded Neural Operators for Elliptic PDEs
Haixin Wang*, Jiaxin Li*, Anubhav Dwivedi, Kentaro Hara, Tailin Wu†
ICLR 2024; Oral at NeurIPS 2023 AI4Science workshop
[paper][arXiv][code]
TL;DR: [ML for simulation] We introduce a boundary-embedded neural operator that incorporates complex boundary shape and inhomogeneous boundary values into the solving of Elliptic PDEs

Uncertainty Quantification for Forward and Inverse Problems of PDEs via Latent Global Evolution
Tailin Wu*, Willie Neiswanger*, Hongtao Zheng*, Stefano Ermon, Jure Leskovec
[paper][arXiv][code]
AAAI 2024 Oral (top 10% of accepted papers)
TL;DR: [ML for simulation] We introduced uncertainty quantification method for forward simulation and inverse problems of PDEs using latent uncertainty propagation.

Learning Controllable Adaptive Simulation for Multi-resolution Physics
Tailin Wu* , Takashi Maruyama*, Qingqing Zhao*, Gordon Wetzstein, Jure Leskovec
ICLR 2023 notable top-25%
Best poster award at SUPR
[paper][project page][arXiv][poster][slides][tweet][code]
TL;DR: [ML for simulation] We introduced the first fully deep learning-based surrogate models for physical simulations that jointly learn forward prediction and optimizes computational cost with RL.

Artificial Intelligence for Science in Quantum, Atomistic, and Continuum Systems
Xuan Zhang, … Tailin Wu (31th/63 author), … Shuiwang Ji
[paper][website]
TL;DR: [ML for simulation] Review paper summarizing the key challenges, current frontiers, and open questions of AI4Science, for quantum, atomistic, and continuum systems.

Learning to Accelerate Partial Differential Equations via Latent Global Evolution
Tailin Wu, Takashi Maruyama, Jure Leskovec
NeurIPS 2022
[project page][arXiv][paper][poster][code]
TL;DR: [ML for simulation] We introduced a method for accelerating forward simulation and inverse optimization of PDEs via latent global evolution, achieving up to 15x speedup while achieving competitive accuracy w.r.t. SOTA models.

Learning Large-scale Subsurface Simulations with a Hybrid Graph Network Simulator
Tailin Wu, Qinchen Wang, Yinan Zhang, Rex Ying, Kaidi Cao, Rok Sosič, Ridwan Jalali, Hassan Hamam, Marko Maucec, Jure Leskovec
SIGKDD 2022 & ICLR AI for Earth and Space Sciences Workshop long contributed talk
[arXiv][project page][publication]
TL;DR: [ML for simulation] We introduced a hybrid GNN-based surrogate model for large-scale fluid simulation, with up to 18x speedup and scale to over 3D, 106 cells per time step (100x higher than prior models).

ZeroC: A Neuro-Symbolic Model for Zero-shot Concept Recognition and Acquisition at Inference Time
Tailin Wu, Megan Tjandrasuwita, Zhengxuan Wu, Xuelin Yang, Kevin Liu, Rok Sosič, Jure Leskovec
NeurIPS 2022
[arXiv][project page][poster][paper][slides][code]
TL;DR: [ML for discovery] We introduce a neuro-symbolic method that trained with simpler concepts and relations, can zero-shot generalize to more complex, hierarchical concepts, and transfer the knowledge across domains.

Learning Efficient Hybrid Particle-continuum Representations of Non-equilibrium N-body Systems
Tailin Wu, Michael Sun, H.G. Jason Chou, Pranay Reddy Samala, Sithipont Cholsaipant, Sophia Kivelson, Jacqueline Yau, Zhitao Ying, E. Paulo Alves, Jure Leskovec†, Frederico Fiuza†
NeurIPS 2022 AI4Science workshop [paper][poster]
TL;DR:[ML for simulation] We introduced a hybrid particle-continuum representation for simulation of multi-scale, non-equilibrium, N-body physical systems, speeding up laser-plasma simulation by 8-fold with 6.8-fold error reduction.

Graph Information Bottleneck
Tailin Wu*, Hongyu Ren*, Pan Li, Jure Leskovec
NeurIPS 2020
[arXiv][project page][poster][slides][code]
Featured in Synced AI Technology & Industry Review (机器之心) [news]
TL;DR: [representation learning] We introduced Graph Information Bottleneck, a principle and representation learning method for learning minimum sufficient information from graph-structured data, significantly improving GNN’s robustness to adversarial attacks and random noise.

Phase transitions for the Information Bottleneck in representation learning
Tailin Wu, Ian Fischer
ICLR 2020
[paper][arXiv][video][NeurIPS ITML][poster]
TL;DR: [representation learning] We theoretically analyzed the Information Bottleneck objective, to understand and predict observed phase transitions (sudden jumps in accuracy) in the prediction vs. compression tradeoff.

Toward an Artificial Intelligence Physicist for Unsupervised Learning
Tailin Wu, Max Tegmark
Physical Review E 100 (3), 033311 [journal article][arXiv][code]
Featured in MIT Technology Review and Motherboard
Featured in PRE Spotlight on Machine Learning in Physics
TL;DR: [ML for discovery] We introduced a paradigm and algorithms for learning theories (small, interpretable models together with domain classifier) each specializing in explaining aspects of a dynamical system. It combines four inductive biases from physicists: divide-and-conquer, Occam’s razor with MDL, unification and lifelong learning.

Discovering Nonlinear Relations with Minimum Predictive Information Regularization
Tailin Wu, Thomas Breuel, Michael Skuhersky, Jan Kautzin
Best Poster Award at ICML 2019 Time Series Workshop
[arXiv][ICML workshop][poster][code]
TL;DR: [ML for discovery] We introduced a method for inferring Granger causal relations for large-scale, nonlinear time series with only observational data.

Pareto-optimal data compression for binary classification tasks
Max Tegmark, Tailin Wu
Entropy 2020, 22(1), 7, as cover issue. arXiv:1908.08961.
[Entropy][arXiv][code]
TL;DR: [representation learning] We introduce an algorithm for discovering the Pareto frontier of compression vs. prediction tradeoff in binary classification of neural networks.

Learnability for the Information Bottleneck
Tailin Wu, Ian Fischer, Isaac Chuang, Max Tegmark
UAI 2019, Tel Aviv, Israel [arXiv][pdf] [poster][slides]
Entropy 21(10), 924 (Extended version) [journal article]
ICLR 2019 Learning with Limited Data workshop as spotlight [openreview][poster]
TL;DR: [representation learning] We theoretically derive the condition of learnability in the compression vs. prediction tradeoff in the Information Bottleneck objective.

AI Feynman 2.0: Pareto-optimal symbolic regression exploiting graph modularity
Silviu-Marian Udrescu, Andrew Tan, Jiahai Feng, Orisvaldo Neto, Tailin Wu, Max Tegmark
NeurIPS 2020, Oral
[arXiv][code]
TL;DR: [ML for discovery] We introduce a state-of-the-art symbolic regression algorithm that robustly re-discovering top 100 physics equations from noisy data from Feynman lectures.

Learning with Confident Examples: Rank Pruning for Robust Classification with Noisy Labels
Curtis G. Northcutt*, Tailin Wu*, Isaac Chuang
UAI 2017
[arXiv][code]
Cleanlab is built on top of it.
TL;DR: [representation learning] We introduce a rank pruning method for classification with noisy labels, which provably obtains similar performance as without label noise.

ViRel: Unsupervised Visual Relations Discovery with Graph-level Analogy
Daniel Zeng*, Tailin Wu*, Jure Leskovec
ICML 2022 Workshop of Beyond Bayes: Paths Towards Universal Reasoning Systems
[arXiv][project page][code]
TL;DR: [ML for discovery] We introduce a method that discovers common relational structures (analogical reasoning) from few-shot examples.

Intelligence, physics and information – the tradeoff between accuracy and simplicity in machine learning
Tailin Wu
PhD Thesis [arXiv]
TL;DR: My PhD thesis summarizing my PhD work on AI for Physics and Physics for AI, together with representation learning with information theory.

Toward a more accurate 3D atlas of C. elegans neurons
Michael Skuhersky, Tailin Wu, Eviatar Yemini, Amin Nejatbakhsh, Edward Boyden, Max Tegmark
BMC bioinformatics 23 (1), 1-18 [paper]
TL;DR: [ML for discovery] We introduce a method for identifying the neuron ID in C. elegans and introduced a more accurate neuron atlas with the NeuroPAL technique.

Meta-learning autoencoders for few-shot prediction
Tailin Wu, John Peurifoy, Isaac L. Chuang, Max Tegmark
arXiv:1807.09912. [arXiv][code]

Preventing and Reversing Vacuum-Induced Optical Losses in High-Finesse Tantalum (V) Oxide Mirror Coatings
Dorian Gangloff*, Molu Shi*, Tailin Wu*, Alexei Bylinskii, Boris Braverman, Michael Gutierrez, Rosanna Nichols, Junru Li, Kai Aichholz, Marko Cetina, Leon Karpa, Branislav Jelenković, Isaac Chuang, and Vladan Vuletić
Optics Express 23(14). (2016) [paper]

Iterative Precision Measurement of Branching Ratios Applied to 5P states in 88Sr+
Helena Zhang, Michael Gutierrez, Guang Hao Low, Richard Rines, Jules Stuart, Tailin Wu, Isaac Chuang
New Journal of Physics, 18(12):123021, (2016) [paper]

Pathway-based mean-field model for Escherichia coli chemotaxis
Guangwei Si, Tailin Wu, Qi Ouyang, and Yuhai Tu
Physical Review Letters, 109, 048101 (2012) [paper]

Frequency-Dependent Escherichia coli Chemotaxis Behavior
Xuejun Zhu, Guangwei Si, Nianpei Deng, Qi Ouyang, Tailin Wu, Zhuoran He, Lili Jiang, Chunxiong Luo, and Yuhai Tu
Physical Review Letters. 108, 128101 (2012) [paper]

Teaching:

  • 2024 Spring: Frontiers in Computer Science and Technology (for PhD students)
  • Guest Lecturer, AI Physicist and Machine Learning for Simulations, for Caltech CS159: Advanced Topics in Machine Learning, April, 2022.
  • Teaching Assistant, MIT 8.01: Classical Mechanics, responsible developing part of the online interactive sessions for providing additional materials and automatic grading, and in-person tutoring, September to December, 2016.
  • Teaching Assistant, MIT 8.02: Electricity and Magnetism, responsible developing part of the online interactive sessions for providing additional materials and automatic grading, and in-person tutoring, February to May, 2017.

Invited talks:

Featured talks:
[2023/03] AI + Science: motivation, advances and open problems, at Swarma (集智) [slides][summary]
[2023/05] Graph Neural Networks for Large-scale Scientific Simulations, at 首届人工智能科学计算研讨会
[2023/03] Steps toward an AI scientist: neuro-symbolic models for concept generalization and theory learning, at AAAI 2023 Symposium of Computational Approaches to Scientific Discovery [slides]
[2023/02] Learning structured representations for accelerating scientific discovery and simulation, at Yale Department of Statistics and Data Science
[2022/06] Learning to Accelerate Large-scale Physical Simulations in Fluid and Plasma Physics, at the Data-driven Physical Simulations (DDPS) seminar at Lawrence Livermore National Laboratory [video]
[2022/04] Guest lecture at Caltech CS159, AI Physicist and Machine Learning for Simulations [slides]

Other invited talks:
[2024/03] Uncertainty Quantification for Forward and Inverse Problems of PDEs via Latent Global Evolution, at SIAM Conference on Uncertainty Quantification (UQ24)
[2024/02] Compositional generative inverse design, at AI4Science Talks [slides]
[2024/01] Machine learning for accelerating scientific simulation, design and control, at Annual Meeting of Peking University Center for Quantitative Biology
[2023/11] GNN for scientific simulations: towards adaptive multi-resolution simulators and a foundation neural operator, at LoG Shanghai Meetup
[2023/08] Machine learning of structured representations for accelerating scientific discovery and simulation, invited talk at CPS Fall Meeting（中国物理学会2023秋季会议）
[2023/08] Graph neural network for accelerating scientific simulation, at 2023科学智能峰会
[2023/08] Phase transitions in the universal tradeoff between accuracy and simplicity in machine learning, at HK Satellite of StatPhys28
[2023/07] 走向AI物理学家，做科学发现和仿真, at 果壳未来光锥
[2023/06] Machine learning of structured representations for accelerating scientific discovery and simulation, at 清华大学
[2023/06] Learning structured representations for accelerating simulation and design, at School of Intelligence Science and Technology, at 北京大学
[2023/05] Steps toward an AI scientist: neuro-symbolic models for concept generalization and theory learning, at Brown University Autonomous Empirical Research group.
[2023/04] AI for scientific design: Surrogate model + backpropagation method, at Swarma [slides]
[2023/04] Learning Controllable Adaptive Simulation for Multi-resolution Physics, at Stanford Data for Sustainability Conference 2023 [slides]
[2023/03] Learning Controllable Adaptive Simulation for Multi-resolution Physics, at TechBeat [slides][video]
[2023/02] Graph Neural Networks for Large-scale Simulations, at Stanford HAI Climate-Centred Student Affinity Group [slides]
[2023/02] ZeroC: A Neuro-Symbolic Model for Zero-shot Concept Recognition and Acquisition at Inference Time, at AI Time [slides]
[2022/12] Steps toward an AI scientist: neuro-symbolic models for zero-shot learning of concepts and theories, at BIGAI
[2022/09] Learning to Accelerate Large-Scale Physical Simulations in Fluid and Plasma Physics, at SIAM 2022 Conference on Mathematics of Data Science
[2022/06] Learning to accelerate simulation and inverse optimization of PDEs via latent global evolution, Stanford CS ML lunch
[2021/12] Machine learning of physics theories, at Alan Turing Institute, Machine Learning and Dynamical Systems Seminar [video]
[2021/10] Graph Information Bottleneck, at Beijing Academy of Artificial Intelligence (BAAI)
[2021/06] Machine learning of physics theories and its universal tradeoff between accuracy and simplicity, at Workshop on Artificial Scientific Discovery 2021 [slides][video]
[2021/04] Machine Learning of Physics Theories, at Workshop on Artificial Scientific Discovery 2021, Summer school hosted by Max Planck Institute for the Science of Light [slides][video]
[2021/04] Phase transitions on the tradeoff between prediction and compression in machine learning, Stanford CS ML lunch
[2021/03] Learning to accelerate the simulation of PDEs, at CLARIPHY Topical Meetings [video]
[2021/02] Machine Learning of Physics Theories, at Seminar Series of SJTU Institute of Science [video]
[2021/02] Phase transitions on the universal tradeoff between prediction and compression in machine learning, at Math Machine Learning seminar MPI MIS + UCLA [video]
[2020/10] Machine learning of physics theories and its universal tradeoff between accuracy and simplicity, at Los Alamos National Lab
[2020/02] Phase transitions for the information bottleneck, UIUC

Reviewers:

  • Proceedings of the National Academy of Sciences (PNAS)
  • Nature Machine Intelligence
  • Nature Communications
  • Nature Scientific Reports
  • Science Advances
  • Patterns (by Cell Press)
  • iScience (by Cell Press)
  • NeurIPS 2023, NeurIPS 2022, NeurIPS 2021, NeurIPS 2020
  • ICML 2024, ICML 2023, ICML 2022, ICML 2021, ICML 2020
  • ICLR 2024, ICLR 2023, ICLR 2022
  • SIGKDD 2023, SIGKDD 2022
  • SIAM 2024 International Conference on Data Mining (SDM’24)
  • Learning on Graph Conference
  • TMLR
  • IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)
  • IEEE Transactions on Image Processing (TIP)
  • IEEE Transactions on Audio, Speech and Language Processing (T-ASL)
  • IEEE Transactions on Network Science and Engineering
  • IEEE Signal Processing Letters
  • IEEE International Symposium on Information Theory (ISIT)
  • Neural Computation
  • MDPI Journals (Entropy, Algorithm, etc.)
  • Physica D
  • Journal of Petroleum Science and Engineering

Education:

  • 2020-2023.04: Stanford University
    Postdoctoral scholar in Computer Science Department
  • 2012-2019: Massachusetts Institute of Technology
    Ph.D. in Department of Physics
  • 2008-2012: Peking University（北京大学）
    Bachelor of Science in Physics

Experience:

Service:

  • Lead organizer and program committee, ICLR 2021 Deep Learning for Simulation Workshop (simDL), 2021.
  • Co-President, MIT Chinese Student and Scholar Association (MIT CSSA), 2015-2016.
  • Executive board member, MIT Chinese Student and Scholar Association (MIT CSSA), 2014-2016
  • Mentor, Stanford CS undergraduate mentoring program for students from underrepresented backgrounds, 2021.
  • Mentor, Stanford summer undergraduate research program (CURIS).

Selected open-source projects:

Alumni (selected from directly supervised students):

My favorites:

  • Video:
    • Timelapse of the future: A Journey to the End of Time: the vastness of the Universe does not lie in its space, but its time. This video is utterly beautiful, exhilarating, and soothing, making us rethink time, and what we want to do in the blink of time the Universe has given us.
  • Sci-fi show:
    • The Expanse: a realistic, amazing sci-fi show depicting the solar system three hundred years from now, where humans have colonized Earth, Mars and the Belt. The story began when they discover proto-molecules from an alien civilization, and a mystery is gradually revealed. Each season expands to a broader scope, finally to a galaxy-spanning human empire.
  • Sci-fi novel:
    • A Fire upon the Deep: a fantastic novel involving multiple levels of super-intelligence, and how some individuals’ actions fundamentally change the future of the galaxy.