| 研究領域 | 社会変革の源泉となる革新的アルゴリズム基盤の創出と体系化 |
|---|
| 研究課題/領域番号 |
23H04377
|
|---|
| 研究種目 |
学術変革領域研究(A)
|
| 配分区分 | 補助金 |
|---|
| 審査区分 |
学術変革領域研究区分(Ⅳ)
|
|---|
| 研究機関 | 東京大学 |
|---|
研究代表者 |
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)
|
|---|
| 研究期間 (年度) |
2023-04-01 – 2025-03-31
|
| 研究課題ステータス |
交付 (2024年度)
|
| 配分額 *注記 |
5,200千円 (直接経費: 4,000千円、間接経費: 1,200千円)
2024年度: 2,600千円 (直接経費: 2,000千円、間接経費: 600千円)
2023年度: 2,600千円 (直接経費: 2,000千円、間接経費: 600千円)
|
|---|
| キーワード | Differential Privacy / Network Optimization / Theoretical Analysis / Graph Algorithm / Graph Optimization / Optimization |
|---|
| 研究開始時の研究の概要 |
We will consider two LDP approaches which are 1) graph stability and 2) two-step publications. Approaches based on graph stability are not scalable, while solutions to the two-step publications are not precise. We aim to speed-up the graph stability approaches and improve the precision of results of the two-step publication. We proposed graph algorithms which are robust against attackers. We will use ideas there to speed up the graph stability approaches. For the approaches based on the two-step publication, some ideas will be from our previous works on star and triangle counting under LDP.
|
| 研究実績の概要 |
We conducted the following research on graph algorithms under local differential privacy.
(1) We complete the research to show that the result of the spectral clustering is not significantly changed even when the graph is obfuscated under the local differential privacy. The result is published at https://arxiv.org/abs/2309.06867.
(2) We present an algorithm that computes the number of paths and Katz centrality while adhering to local differential privacy standards. This work is among the first to incorporate global graph information into local differential privacy frameworks. The result is published at https://arxiv.org/abs/2310.14000. Additionally, the paper has been accepted through a peer review process and will be featured in the proceedings of UAI 2024.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
We have conducted a research based on the plan, and have published the result at the refereed conference proceeding.
|
| 今後の研究の推進方策 |
In fiscal year 2024, we have a plan to conduct the following two works:
(1) We will develop local clustering algorithms under local differential privacy. Compared to the spectral clustering algorithm which we have analyze in fiscal year 2023, we believe that we can obtain a better result when focusing on the task of local clustering.
(2) We will continue our work on subgraph counting problems. Specifically, we will work on the graph with small arboricity. While most of the current works focus on triangle counting, we plan to also work on the counting of larger subgraphs.
|
