2021 Fiscal Year Annual Research Report
Precision Analysis of Frameworks for Publishing Graph Information under Differential Privacy
Publicly Offered Research
| Project Area | Creation and Organization of Innovative Algorithmic Foundations for Leading Social Innovations |
|---|
| Project/Area Number |
21H05845
|
|---|
| Research Institution | The University of Tokyo |
|---|
Principal Investigator |
スッパキットパイサン ウォラポン 東京大学, 大学院情報理工学系研究科, 特任准教授 (30774103)
|
|---|
| Project Period (FY) |
2021-09-10 – 2023-03-31
|
| Keywords | differential privacy / Graph metrics / Spectral Clustering / Isogeny-based crypto |
|---|
| Outline of Annual Research Achievements |
During the course of this support, we conducted various theoretical and experimental analyses on differential privacy, which involves partially corrupting users' information to protect their privacy. Our research focused on analyzing the robustness of different algorithms against this corruption, and we made two significant contributions in this area.
(A) We analyzed the precision of publishing graph statistics under local differential privacy and discovered a significant bias in the publication. To address this issue, we proposed a method to completely eliminate the bias, resulting in up to 100 times better precision compared to the state-of-the-art algorithm when publishing the number of k-stars and the variance of degrees.
(B) We demonstrated that machine learning algorithms using symmetric loss are robust against corruption under the randomized response and exponential mechanism.
In addition to our work on differential privacy, we also made contributions to the Isogeny-based cryptographic system. We developed an algorithm that speeds up the system when implemented in parallel environments, such as multi-core CPUs or SIMD. Our runtime is up to 20% smaller than the state-of-the-art algorithm based on scheduling algorithms. We also optimized a SIMD architecture called Support Vector Extension (SVE), increasing the throughput of the cryptographic protocol by nearly four times.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
Our progress in the field of differential privacy is on track, as we have established a foundation for research in the fiscal year 2022. We have also published a paper on our contributions, as outlined in point (B) in the "Summary of Research Achievements," at the international workshop PDAA 2022.
Additionally, we were able to utilize several ideas from differential privacy to accelerate the calculation of the isogeny-based cryptographic system. Although this contribution was not anticipated at the outset of the project, it resulted in several advancements to the cryptosystem. We have published papers on this topic at both ACISP 2022 and CANDAR 2022.
|
| Strategy for Future Research Activity |
In the fiscal year 2022, we utilized the insights gained from our research on both differential privacy and Isogeny-based cryptography.
Regarding our work on differential privacy, we intend to continue improving our algorithm for removing bias resulting from local differential privacy. We plan to apply this algorithm to other combinatorial problems such as k-clique or graph partition problem in the coming fiscal year.
As for our research on Isogeny-based cryptography, we plan to implement our scheduling algorithm in real hardware. Additionally, while we achieved an algorithm with high throughput in the previous academic year through optimization on SIMD architecture, we aim to extend these ideas further to develop an algorithm with small latency in fiscal year 2022.
|
