So Takao

Postdoctoral Reseach Scholar @ Caltech

Education

PhD in Applied Mathematics

Imperial College London
2016 - 2020

MSc in Applied Mathematics (Pass with Distinction, 87/100)

Imperial College London
2015 - 2016

BSc in Mathematics (First class honour, 81/100)

Imperial College London
2012 - 2015

Career Profile

So Takao is a postdoctoral scholar at the California Institute of Technology, working on the intersection of machine learning and data assimilation. His interests involve (1) developing efficient inference techniques for probabilistic models such as Gaussian processes and stochastic partial differential equations, (2) developing machine learning models with physical, geometric or topological inductive biases, and (3) applying machine learning techniques to problems in weather or climate science. During 2021-2023, he was a Senior Research Fellow in Machine Learning for Climate Science at the UCL Sustainability and Machine Learning group. There, he lead the Met Office Academic Partnership workgroup on “Data science methodology for weather and climate” and intiated various collaborations across departments to work on problems related to climate change. He received his PhD in 2020 from the Department of Mathematics at Imperial College London, where he wrote his thesis on structure-preserving fluid models and MCMC techniques on Lie groups, advised by Prof. Darryl Holm.

Experiences

Postdoctoral Scholar Research Associate

2024 - Present
California Institute of Technology
  • Principle investigator: Prof. Andrew Stuart
  • Working on various projects involving Gaussian processes, data assimilation, uncertainty quantification and operator learning.

Research Fellow in Machine Learning for Climate Science

2021 - 2023
University College London
  • Principle investigator: Prof. Marc Deisenroth
  • Worked on developing novel data assimilation algorithms using ideas from message passing, Gaussian markov random fields, stochastic (partial) differential equations, differential geometry and algebraic topology.
  • Collaborated with the UCL Earth Science Department on various projects on sea-ice modelling.
  • Supervised several BSc and MSc students.
  • Lead the Met Office Academic Partnership workgroup on “Data science methodology for weather and climate”.

Preprints and Publications

Preprints

  • Semimartingale driven mechanics and reduction by symmetry for stochastic and dissipative dynamical systems
    • Oliver Street, So Takao
    arXiv:2312.09769 (2023)
  • Transport noise restores uniqueness and prevents blow-up in geometric transport equations
    • Aythami Bethencourt de León, So Takao
    arXiv:2211.14695 (2022)
  • A unifying and canonical description of measure-preserving diffusions
    • Alessandro Barp, So Takao, Michael Bethencourt, Alexis Arnaudon, Mark Girolami
    arXiv:2105.02845 (2021)

    Publications

  • Scalable interpolation of satellite altimetry data with probabilistic machine learning
    • William Gregory, Ronald MacEachern, So Takao, Isobel Lawrence, Carmen Nab, Marc Deisenroth, Michel Tsamados
    Nature Communications (2024)
  • ARISGAN Extreme Super-Resolution of Arctic Surface Imagery using Generative Adversarial Networks
    • Christian Au, Michel Tsamados, Petru Manescu, So Takao
    Frontiers in Remote Sensing (2024)
  • Co-located OLCI optical imagery and SAR altimetry from Sentinel-3 for enhanced Arctic spring sea ice surface classification
    • Weibin Chen, Michel Tsamados, Rosemary Willatt, So Takao, others
    Frontiers in Remote Sensing (2024)
  • Scalable data assimilation with message passing
    • Oscar Key, So Takao, Daniel Giles, Marc Peter Deisenroth
    Climate Informatics (2024)
  • Gaussian processes on cellular complexes
    • Mathieu Alain, So Takao, Brooks Paige, Marc Deisenroth
    International Conference on Machine Learning (2024)
  • Iterated INLA for state and parameter estimation in nonlinear dynamical systems
    • Rafael Anderka, Marc Peter Deisenroth, So Takao
    Uncertainty in Artificial Intelligence (2024)
  • A geometric extension of the Itô-Wentzell and Kunita's Formulas
    • Aythami Bethencourt de León, So Takao
    Stochastic Processes and Their Applications (2024)
  • Actually sparse variational Gaussian processes
    • Harry Jake Cunningham, Daniel Augusto de Souza, So Takao, Mark van der Wilk, Marc Deisenroth
    International Conference on Artificial Intelligence and Statistics (2023)
  • Iterative state estimation in non-linear dynamical systems using approximate expectation propagation
    • Sanket Kamthe, So Takao, Shakir Mohamed, Marc Deisenroth
    Transactions on Machine Learning Research (2022)
  • Vector-valued Gaussian processes on Riemannian manifolds via gauge-equivariant projected kernels
    • Michael Hutchinson, Alexander Terenin, Viacheslav Borovitskiy, So Takao, Yee Whye Teh, Marc Deisenroth
    Neural Information Processing Systems (2021)
  • Irreversible Langevin MCMC on Lie Groups
    • Alexis Arnaudon, Alessandro Barp, So Takao
    Geometric Science of Information (2019)
  • Modelling the climate and weather of a 2D Lagrangian-averaged Euler-Boussinesq equation with transport noise
    • Diego Alonso-Orán, Aythami Bethencourt de León, Darryl Holm, So Takao
    Journal of Statistical Physics (2019)
  • Implications of Kunita-Itô-Wentzell formula for k-forms in stochastic fluid dynamics
    • Aythami Bethencourt de León, Darryl Holm, Erwin Luesink, So Takao
    Journal of Nonlinear Science (2019)
  • The Burgers' equation with stochastic transport: shock formation, local and global existence of smooth solutions
    • Diego Alonso-Orán, Aythami Bethencourt de León, So Takao
    Nonlinear Differential Equations and Applications (2019)
  • Impacts of atmospheric reanalysis uncertainty on Atlantic Overturning Estimates at 25°N
    • Helen R. Pillar, Helen L. Johnson, David P. Marshall, Patrick Heimbach, So Takao
    Journal of Climate (2018)
  • Networks of Coadjoint Orbits: from geometric to statistical mechanics
    • Alexis Arnaudon, So Takao
    Journal of Geometric Mechanics (2018)

    Workshop papers

  • Short-term prediction and filtering of solar power using state-space Gaussian processes
    • Sean Nassimiha, Peter Dudfield, Jack Kelly, Marc Deisenroth, So Takao
    NeurIPS workshop on Tackling Climate Change with Machine Learning (2022)
  • Actually sparse variational Gaussian processes
    • Jake Cunningham, So Takao, Mark van der Wilk, Marc Deisenroth
    NeurIPS workshop on Gaussian Processes, Spatiotemporal Modelling, and Decision-Making Systems (2022)

    OSS Contributions

    Below is a list of open source softwares that I have been involved in developing:

    GPSat - Python package to interpolate nonstationary geospatial fields from observation data using local Gaussian process models
    EPyStateEstimator - Iterative state estimation in non-linear dynamical systems using approximate expectation propagation
    ExtrinsicGaugeIndependentVectorGPs - A library implementing the kernels for and experiments using extrinsic gauge equivariant vector field Gaussian Processes

    Invited Talks

  • The Euler Equations. A Coincidence or Genius?
    • SIAM-JAMS (2017)
  • Networks of Coadjoint Orbits. Bridging the gap between geometric and statistical mechanics.
    • Imperial SIAM Student Chapter Annual Conference (2018)
  • Modelling Uncertainty in the Ocean. A Geometric Perspective.
    • SIAM-JAMS (2019)
  • Extending the Generalised HMC to Lie Groups and Beyond.
    • Geometric Science of Information Conference (2019)
  • Geometric Framework for Stochastic GFD Modelling.
    • Nonlinear and Stochastic Methods in Climate and GFD, Climate and GFD, Institute Henri Poincare (2019)
  • Generalised Hamiltonian Monte Carlo on Lie Groups.
    • SIAM-JAMS (2019)
  • Stochastic Advection by Lie Transport. The Past, Present and Future.
    • Second Applied Geometric Mechanics Meeting on “Stochastic Geometric Mechanics, Fluid Models and Uncertainty Quantification” (2019)
  • Machine-learned 4DVar. A case study with the L63 Model.
    • STUOD Inaugural Workshop (2020)
  • Intelligent Weather Prediction. Can A.I. be used to produced better forecasts?
    • UCL ComputerScience PhD Seminar (2020)
  • Vector-valued Gaussian Processes on Manifolds.
    • Met Office Academic Partnership Workshop on Uncertainty Quantification and Parameterization (2021)
  • Spherical models for data driven weather forecasting.
    • UCL Statistical Machine Learning Internal Seminar (2021)
  • A novel framework for data assimilation using message passing.
    • UCL ML4Climate (2022)
  • Incorporating physics into spatiotemporal message passing.
    • UCL Statistical Machine Learning Research Day (2022)
  • Rethinking Data Assimilation as a Message Passing Problem.
    • UCL Statistical Machine Learning Internal Seminar (2022)
  • Improving data-assimilation for weather forecasting. A graph-based Bayesian perspective.
    • RIKEN, Tokyo (2023)
  • Data Assimilation. A message passing perspective.
    • CliMA, Caltech (2024)
  • Data Assimilation. A message passing perspective.
    • SIAM talk Caltech (2024)

    Professional Roles

    Organised workshops

  • UCL Met Office Academic Partnership sandpit meeting on “Uncertainty quantification and parameterizations” (2021)
  • UCL Met Office Academic Partnership workshop on Bayesian machine learning for weather and climate (2022)
  • UCL AI Centre workshop on “AI for Sustainability” (2023)

    • Teaching

  • Machine Learning Seminar course on “Message Passing Algorithms in Machine Learning”, UCL (2023)

    • Supervision

  • Nanxi Zhang
    • Solar PV Nowcasting with graph neural networks (2021)
    with M. Deisenroth and J. Kelly
  • Rui Li
    • Learning input conditional invariances using the marginal likelihood (2021)
    with M. Deisenroth and M. van der Wilk
  • Eiki Shimizu
    • Improving the approximate inference of invariant Gaussian processes (2021)
    with M. Deisenroth and M. van der Wilk
  • Sean Nassimiha
    • Modelling solar power production with spatio-temporal variational Gaussian processes (2022)
    with M. Deisenroth and P. Dudfield
  • Ronald Maceachern
    • Sea ice freeboard optimal interpolation (2022)
    with M. Deisenroth, M. Tsamados and W. Gregory
  • Bengt Lofgren
    • Boundary aware Gaussian processes. treading freely near the edge. (2022)
    with M. Deisenroth and J. Cunningham
  • Weibin Chen
    • Co-located OLCI optical imagery and SAR altimetry from Sentinel-3 for enhanced Arctic spring sea ice surface classification
    with M. Tsamados and R. Willatt
  • Eirik Aalstad Baekkelund
    • Probabilistic Solar PV Nowcasting (2023)
    with M. Deisenroth
  • Rafael Anderka
    • Efficient Data Assimilation With Nonlinear Stochastic Partial Differential Equations Through Markov Structures (2023)
    with M. Deisenroth
  • Christian Au
    • ARISGAN Extreme Super-Resolution of Arctic Surface Imagery using Generative Adversarial Networks (2023)
    with M. Tsamados and P. Manescu

    Scholarships and Awards

  • Schrödinger Scholarship Scheme for Mathematics (2016-2020)
  • Imperial SIAM Student Chapter Annual Conference. Best presentation award (2018).
  • Doris Chen Mobility Award (2018-2019)

    • Other Roles

  • Reviewed papers for top tier academic journals and conferences including, Annals of Applied Probability (AAP), International Conference on Machine Learning, Conference on Neural Processing Systems (NeurIPS), International Conference on Learning Representations (ICLR) and Journal of Machine Learning Research (JMLR).
  • Research lead for the Met Office Academic Partnership (MOAP) work group on “Applications of Data Science to Weather and Climate” (2021-2023)