Donghwan Rho

Hello. I am a Ph.D. candidate in Department of Mathematical Sciences, Seoul National University under the supervision of Ernest K. Ryu.

I am interested in:

Efficient implementations of language models under Homomorphic Encryption (especially CKKS)
Mathematical reasoning of LLMs
Interpretability of language models

GitHub / Google Scholar / LinkedIn

Feel free to contact me anytime to discuss research!

News

Jan 23, 2025	Our first paper ‘Encryption-Friendly LLM Architecture’ is accepted at ICLR’25.

Publications

Preprint

Homomorphically Encrypted Language Models Stabilized with TSP-Sorted Tokens

D. Rho^*, S. Seo^*, H. Sung, and 2 more authors

2025

Bib

@article{rho2025ckks_randomsampling,
  author = {Rho, D. and Seo, S. and Sung, H. and Min, C. and Ryu, E. K.},
  title = {Homomorphically Encrypted Language Models Stabilized with TSP-Sorted Tokens},
  year = {2025},
}

ICLR
Encryption-Friendly LLM Architecture

D. Rho^*, T. Kim^*, M. Park, and 4 more authors

International Conference on Learning Representations, 2025

Abstract Bib HTML Code

Large language models (LLMs) offer personalized responses based on user interactions, but this use case raises serious privacy concerns. Homomorphic encryption (HE) is a cryptographic protocol supporting arithmetic computations in encrypted states and provides a potential solution for privacy-preserving machine learning (PPML). However, the computational intensity of transformers poses challenges for applying HE to LLMs. In this work, we propose a modified HE-friendly transformer architecture with an emphasis on inference following personalized (private) fine-tuning. Utilizing LoRA fine-tuning and Gaussian kernels, we achieve significant computational speedups—6.94x for fine-tuning and 2.3x for inference—while maintaining performance comparable to plaintext models. Our findings provide a viable proof of concept for offering privacy-preserving LLM services in areas where data protection is crucial.
@article{rho2025iclr, title = {Encryption-Friendly LLM Architecture}, author = {Rho, D. and Kim, T. and Park, M. and Kim, J. W. and Chae, H. and Ryu, E. K. and Cheon, J. H.}, journal = {International Conference on Learning Representations}, year = {2025}, url = {https://openreview.net/forum?id=pbre0HKsfE}, }