Tae-Geun Kim-Morgen
Postdoctoral Researcher · Publications under Tae-Geun Kim
Key Laboratory of Nuclear Physics and Ion-beam Application (MOE), Fudan University
RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences (iTHEMS)
Dark matter phenomenology, AI for Science, and Science of AI
GitHub Scholar ORCID CV LinkedIn Email
7
Publications
28
Software
4
Posts
Research
Trustworthy neural samplers
When can an autoregressive neural sampler be trusted for lattice spin systems? I study how much context a sampler needs given the structure of the interaction graph, and how to diagnose failures without exact reference data.
Inverse problems and degeneracy
Which parts of a physical model can data actually determine? With neural operators and identifiability analysis, I ask what gamma-ray spectra reveal about primordial black hole mass distributions, and which ultraviolet couplings SMEFT coefficients pin down.
Scientific machine learning and open-source tools
Operator learning for Hamiltonian dynamics, learning-rate schedules that stay robust when the training budget changes, and numerical software for physicists, including the Rust library Peroxide.
News
  • 2026.11Parallel talk at ATHIC 2026, Shanghai
  • 2026.09Contributed talk at the ECT* workshop, Trento
  • 2026.09Invited seminar at Bielefeld University
  • 2026.08Invited ITP seminar at Heidelberg University; talk and poster at EuCAIFCon 2026
  • 2026.08Invited talk at CCP2026, Seoul, with the APS DCOMP Travel/Registration Award
  • 2026.08Peroxide paper published in the Journal of Open Source Software
  • 2026.07Invited seminars at KIAS and PARTIQAL @ IBS
  • 2026.01PBH axion paper published in Progress of Theoretical and Experimental Physics
  • 2025.11Joined the Fudan–RIKEN joint postdoctoral program (Fudan University & RIKEN iTHEMS)
Recent Publications
Peroxide: A Batteries-Included Numerical Computing Library for Rust
2026
J. Open Source Softw.
Tae-Geun Kim, Giorgio Comitini, Jonas Grage, Benjamin Joens, Marc Schreiber, Soumya Sen, Russell R P Senthamarai, Johanna Sörngård
Scientific ComputingRustOpen Source Software
Abstract: Peroxide is a Rust numerical computing library that integrates eight domains in one crate: linear algebra, ordinary differential equation (ODE) solving, adaptive quadrature, spline interpolation with exact polynomial calculus, root-finding, forward-mode automatic differentiation (AD), statistics, and multi-format DataFrame I/O. We designed Peroxide for applied scientists, engineers, and students who need a cohesive numerical environment in Rust without assembling a multi-crate dependency graph.
@article{kim2026peroxide, title={Peroxide: A Batteries-Included Numerical Computing Library for Rust}, author={Kim, Tae-Geun and Comitini, Giorgio and Grage, Jonas and Joens, Benjamin and Schreiber, Marc and Sen, Soumya and Senthamarai, Russell R P and S{\"o}rng{\aa}rd, Johanna}, journal={Journal of Open Source Software}, volume={11}, number={124}, pages={10366}, year={2026}, doi={10.21105/joss.10366} }
DOI Code
Primordial Black Holes as a Factory of Axions: Extragalactic Photons from Axions
2026
Prog. Theor. Exp. Phys.
Yongsoo Jho*, Tae-Geun Kim*, Jong-Chul Park*, Seong Chan Park*, Yeji Park*
Dark MatterPBHAxion
Abstract: Primordial black holes (PBHs) are significant sources of axions and axion-like particles, provided their Hawking temperature exceeds the particles’ masses. Given the predominant decay of axions into photons, the enhanced photon spectrum they generate can be feasibly detected using sensitive detectors. This paper introduces a novel methodology that elucidates the decay process by which particles traverse and decay over cosmological timescales.
@article{jho2026primordial, title={Primordial Black Holes as a Factory of Axions: Extragalactic Photons from Axions}, author={Jho, Yongsoo and Kim, Tae-Geun and Park, Jong-Chul and Park, Seong Chan and Park, Yeji}, journal={Progress of Theoretical and Experimental Physics}, volume={2026}, number={2}, pages={023B07}, year={2026}, doi={10.1093/ptep/ptag011} }
DOI arXiv
Learning Unknown Gravitational Fields for Spacecraft State Estimation: Temporal Hamiltonian Neural Network-based Adaptive Kalman Filtering
2025
Under review at IEEE TAES
Taehyeun Kim*, Tae-Geun Kim*, Anouck Girard, Ilya Kolmanovsky
Machine LearningHamiltonianBayesian
Abstract: Earlier version on arXiv as "Learning Hamiltonian Dynamics with Bayesian Data Assimilation": In this paper, we develop a neural network-based approach for time-series prediction in unknown Hamiltonian dynamical systems. Our approach leverages a surrogate model and learns the system dynamics using generalized coordinates (positions) and their conjugate momenta while preserving a constant Hamiltonian. Additionally, we employ Bayesian data assimilation to refine predictions in real-time using online measurement data.
@article{kim2025learning, title={Learning Hamiltonian Dynamics with Bayesian Data Assimilation}, author={Kim, Taehyeun and Kim, Tae-Geun and Girard, Anouck and Kolmanovsky, Ilya}, journal={arXiv preprint arXiv:2501.18808}, year={2025} }
arXiv
* Co-first authors · All publications
Recent Talks
Interaction Geometry and Hidden Bias in Learning Lattice Spin Systems
2026
ATHIC 2026 (parallel talk)
Shanghai, China
Trustworthy Neural Sampler
2026
ECT* Workshop
Trento, Italy
Can one see the mass of a PBH?
2026
Seminar @ Bielefeld University · Invited
Bielefeld, Germany
Can you trust your neural sampler?
2026
ITP Seminar @ Heidelberg University · Invited
Heidelberg, Germany
All talks
Featured Software
Peroxide
Comprehensive Rust numerical computing library
Rust ★ 733 ↧ 1.4M
NumericLinear AlgebraStatisticsODE
Comprehensive numerical computing library for Rust, providing functionality comparable to NumPy/SciPy. Core infrastructure for scientific computing research. Published in the Journal of Open Source Software (JOSS), 11(124), 10366.
  • Linear algebra with BLAS/LAPACK integration
  • Optimization algorithms (Gradient Descent, Levenberg-Marquardt)
  • Numerical integration & ODE/PDE solvers
  • Statistical distributions & special functions
  • DataFrame with multiple I/O formats
GitHub Docs
Neural Hamilton
Operator learning for Hamiltonian mechanics
Python ★ 14
Operator LearningHamiltonianNeural ODE
Official implementation of operator learning for Hamiltonian mechanics. Explores whether AI can truly understand physical dynamics.
  • Four neural architectures (DeepONet, TraONet, VaRONet, MambONet)
  • Novel potential generation via Gaussian Random Fields
  • Multi-language (Python, Rust, Julia)
GitHub
Puruspe
Pure Rust special functions library
Rust ★ 29 ↧ 1.6M
Special FunctionsMathematics
Pure Rust implementation of mathematical special functions for scientific computing.
  • Gamma, Beta, Error functions
  • Regularized and inverse variants
  • No external dependencies
GitHub Docs
Honors & Fellowships
APS DCOMP Travel/Registration Award
2026
APS Division of Computational Physics, CCP2026
Shanghai Superpostdoc Fellowship
2025–2027
Shanghai Municipal Government
Fudan Superpostdoc Fellowship
2025–2027
Fudan University
Academy Research Fellowship
2022–2023
Yonsei University
Best Oral Presentation Award
2022
KPS 70th Anniversary and 2022 Fall Meeting