Joon-Woo Lee

Timeline

1. June 2021

   ResNet-20 on encrypted images

   In June 2021, he is first author, with ten co-authors including Yongwoo Lee, Eunsang Lee and Jong-Seon No, of a paper that implements the standard ResNet-20 on the RNS-CKKS homomorphic encryption scheme with bootstrapping. The encrypted model classifies CIFAR-10 images with 90.67% accuracy and gives 98.67% identical results to the original ResNet-20 on unencrypted data.

2. October 2021

   High-precision bootstrapping at Eurocrypt

   As a PhD student on Jong-Seon No's team at Seoul National University, he is first author of the high-precision bootstrapping algorithm for CKKS homomorphic encryption that the team presents at Eurocrypt 2021 in October 2021, with Eunsang Lee, Yongwoo Lee and Young-Sik Kim. The method raises the accuracy of ciphertext computation by up to 1,000 times and, in collaboration with Samsung Advanced Institute of Technology, supports ResNet-20 classification of homomorphically encrypted images.

3. 2022

   Multiplexed parallel convolutions

   In 2022, he co-authors a paper at the 39th International Conference on Machine Learning with Eunsang Lee, Junghyun Lee, Young-Sik Kim, Yongjune Kim, Jong-Seon No and Woosuk Choi. Multiplexed parallel convolutions give ResNet-20 on RNS-CKKS 4.67x lower inference latency and 134x less amortized runtime than earlier work, and the team runs ResNet-110 on RNS-CKKS for the first time.

4. July 2022

   A talk at KAIST

   In July 2022, while a doctoral candidate in electrical and computer engineering at Seoul National University, he gives a KAIST seminar titled "Privacy-Preserving Convolutional Deep Neural Networks on the Fully Homomorphic Encryption". The talk covers CKKS and RNS-CKKS with high-precision bootstrapping and privacy-preserving ResNet and VGGNet on CIFAR and ImageNet. He holds a B.S. in electrical and computer engineering from Seoul National University.

5. August 2022

   Doctorate from Seoul National University

   He completes his PhD in the Department of Electrical and Computer Engineering at Seoul National University in August 2022, advised by Jong-Seon No. His graduate work centers on primitives in fully homomorphic encryption, advanced FHE algorithms and efficient privacy-preserving machine learning with deep neural networks.

6. Since September 2022

   Assistant professor at Chung-Ang

   In September 2022, he joins Chung-Ang University as an assistant professor in the School of Computer Science and Engineering, with research on privacy-preserving machine learning using cryptographic tools. In 2024, he and Joohee Lee write Homomorphic Encryption with Authority, a framework for CKKS-based privacy-preserving services that adds a legally authorized authority party.

7. 2025

   Powerformer at ACL 2025

   In 2025, he co-authors Powerformer with Dongjin Park and Eunsang Lee, a homomorphic-encryption language model published in the proceedings of ACL 2025. Its BERT-base version cuts computation time by 45% against the previous best HE-based private language model, with a distillation step that swaps softmax and layer normalization for power and linear functions.