Hello, I'm Tae-Geun Kim 👋

🙋‍‍♂️ Introduce myself

👨‍‍🏫 Postdoctoral Researcher at Physics

• Institute of Modern Physics, Fudan University & RIKEN iTHEMS
• Curriculum Vitae
• Blog

❤️ Interests

• High energy astrophysics, dark matter and cosmology
• Scientific computation
• Machine Learning / Deep Learning / Statistics
• Quantum Computing

▶️ Status

💼 Portfolio

Peroxide

• Rust numeric library for linear algebra, numerical analysis, statistics, and machine learning
• Provides customizable features for pure Rust, BLAS/LAPACK integration, plotting, and data handling
• Offers user-friendly syntax similar to R, NumPy, and MATLAB
• Supports functional programming, automatic differentiation, and various numerical algorithms
• Includes statistics, special functions, plotting, and DataFrame capabilities
• Compatible with mathematical structures and leverages Rust's performance and package management

HyperbolicLR

• Novel learning rate schedulers (HyperbolicLR and ExpHyperbolicLR) for deep learning optimization
• Addresses the learning curve decoupling problem observed in conventional schedulers
• Demonstrates consistent performance improvements and stability across increasing epoch numbers
• Shows superior performance in maintaining stable learning curves as training duration increases
• Exhibits versatility across various deep learning tasks and model architectures
• Offers potential for more efficient hyperparameter tuning in long-duration training tasks
• Provides a promising approach to improving the training of deep neural networks
• Implemented and evaluated using PyTorch, with experiments on image classification, time series prediction, and operator learning tasks

Puruspe

• Pure Rust library for special functions with no dependencies
• Implements gamma, beta, and error functions
• Provides regularized and inverse versions of the functions
• Lightweight and efficient implementation
• Ideal for mathematical and scientific computing applications
• Based on algorithms from "Numerical Recipes" by Press and Vetterling

Forger

• Reinforcement Learning (RL) library in Rust
• Modular design with components for agents, environments, policies, and utilities
• Efficient and safe implementation leveraging Rust's performance and safety features
• Provides a framework for creating and managing diverse RL environments
• Supports customizable agent strategies and learning algorithms
• Includes implementations of Epsilon Greedy Policy, Value Iteration, and Q-Learning

📚 Publications

• Taehyeun Kim, Tae-Geun Kim, Anouk Girard, Ilya Kolmanovsky, Learning Hamiltonian Dynamics with Bayesian Data Assimilation, arXiv:2501.18808 (2025)

• Tae-Geun Kim, Seong Chan Park, Neural Hamilton: Can A.I. Understand Hamiltonian Mechanics?, arXiv:2410.20951 (2024)

• Tae-Geun Kim, HyperbolicLR: Epoch insensitive learning rate scheduler, arXiv:2407.15200 (2024)

• Chang Min Hyun, Tae-Geun Kim, and Kyounghun Lee, Unsupervised sequence-to-sequence learning for automatic signal quality assessment in multi-channel electrical impedance-based hemodynamic monitoring, CMPB 108079, arXiv:2305.09368 (2023)

• Kayoung Ban, Dong Woo Kang, Tae-Geun Kim, Seong Chan Park and Yeji Park, DeeLeMa : Missing information search with Deep Learning for Mass estimation, Phys. Rev. Research 5, 043186, arXiv:2212.12836 (2022)

• Yongsoo Jho, Tae-Geun Kim, Jong-Chul Park, Seong Chan Park and Yeji Park, Axions from Primordial Black Holes, arXiv:2212.11977 (2022)

:octocat: Github contributions

🏆 Trophies