Efficient Homomorphic Matrix Computation for Secure Transformer Inference w/ Miran Kim

​#Abstract

​Homomorphic encryption (HE) enables computation directly on encrypted data without requiring decryption, thereby preserving data confidentiality. This talk introduces a new diagonal-encoding–based method for homomorphic matrix multiplication. Building on this technique, I present THOR, a secure transformer inference framework that designs core transformer operations to achieve both high efficiency and numerical stability in the encrypted domain. I conclude by presenting comprehensive benchmark results that demonstrate practical end-to-end secure inference for transformer models on encrypted data, highlighting significant performance improvements over prior HE-based approaches.

​#About the Speaker

Miran Kim is an associate professor of the Department of Mathematics and affiliated with the Department of Computer Science at Hanyang University. She received her Ph.D. in mathematical sciences from Seoul National University, Korea, in 2017. Her research focuses on the design of novel strategies for secure and privacy-preserving data analysis using homomorphic encryption. She has extensive experience in the algorithmic optimization of homomorphic encryption, as well as in the design and implementation of efficient protocols for data query processing, genomic data analysis, and machine learning.

​#More Information

Want more information, recordings of the presentations, slides, and other resources from the meetup? Visit this meetup's resource page at https://fhe.org/meetups/084

#Never Miss an Update

The newsletter where we post community announcements: https://fheorg.substack.com/

​The discord server where you can discuss FHE related topics with the community: https://discord.fhe.org

​Make sure to join either (or both) of these to stay informed about future events!