Privacy-Preserving AI
Efficient and practical encrypted inference for Transformers and large language models.
ABOUT ME
Advancing Privacy-Preserving AI
through Efficient Encrypted Inference.
I am a Ph.D. student in the Department of Mathematics at Ewha Womans University, specializing in homomorphic encryption and privacy-preserving AI.
My research focuses on enabling efficient and practical privacy-preserving inference for Transformers and large language models (LLMs). I work on optimizing encrypted computation through algorithmic design, mathematical analysis, and high-performance C++/CUDA implementations, bridging the gap between cryptographic theory and real-world AI applications.
Homomorphic EncryptionCKKSEncrypted LLM InferenceGPU Acceleration
WHAT I WORK ON
Research
Homomorphic Encryption
Algorithmic optimization of CKKS to improve the efficiency and reliability of encrypted computation.
GPU Acceleration
High-performance C++/CUDA implementations for scalable privacy-preserving AI systems.
RECENT UPDATES
News
Error Propagation-Aware Scale Design is available as an IACR ePrint preprint.
Zephyr, our GPU-efficient homomorphic encryption work for Transformer inference, is available online with public code.
Two papers on CKKS bootstrapping and key switching were published in IEEE Access.
SELECTED WORK
Publications & Preprints
Error Propagation-Aware Scale Design for Efficient Homomorphic Encryption-Based LLM Inference
IACR ePrint 2026/2242 · Under review at ICLR 2027
Error-aware scale optimization for efficient encrypted language model inference.
Traveling Salesman-Based Token Ordering Improves Stability in Homomorphically Encrypted Language Models
arXiv:2510.12343 · Co-first author
SIMD-compatible sampling and token ordering for stable encrypted text generation.
Optimal Design of Key Switching Parameters for Efficient CKKS Bootstrapping
IEEE Access, vol. 13, pp. 163431–163446
Key switching parameter optimization to increase available computation depth.
EvalRound+ Bootstrapping and Its Rigorous Analysis for CKKS Scheme
IEEE Access, vol. 13, pp. 140847–140866
Rigorous error analysis and improved modulus efficiency for CKKS bootstrapping.
BACKGROUND
Experience & Education
2021 — Present
Graduate Researcher
Cryptography Research Lab · Ewha Womans University
Research and implementation of CKKS bootstrapping, key switching, GPU acceleration, and privacy-preserving Transformer inference.
2021 — 2027*
Integrated M.S./Ph.D. in Mathematics
Ewha Womans University
*Ph.D. degree expected August 2027.
2017 — 2021
B.S. in Mathematics
Ewha Womans University · Minor in Computer Science
TECHNICAL SKILLS
C++ · CUDA · Python · PyTorch · CKKS / FHE · GPU Optimization · Linux · Git
GET IN TOUCH
Contact
I welcome research discussions and collaborations in cryptography and privacy-preserving AI.
Emailhello0x7@naver.com
Email opens a Gmail compose window. You can also copy the address above.