| 研究課題/領域番号 |
22H03573
|
|---|
| 配分区分 | 補助金 |
|---|
| 研究機関 | 東京大学 |
|---|
研究代表者 |
FAN ZIPEI 東京大学, 空間情報科学研究センター, 客員研究員 (70835397)
|
|---|
| 研究期間 (年度) |
2022-04-01 – 2025-03-31
|
| キーワード | Crowdsensing / Graph Neural Network / Privacy Preserving / Data Valuation |
|---|
| 研究実績の概要 |
In this fiscal year, I conducted the research relevant to machine unlearning which is like federated learning but more recent method that preserves the crowdsensing participants privacy. In this setting, the user’s contribution can be retracted from the trained model. A study on the privacy leakage risk study has been published in GLOBECOM 2023.
Moreover, in the direction of data valuation, I had a deep study on causality that estimating the true effect of each factor on the outcome. Under this framework, more accurate and mathematical guaranteed valuation will be designed. One study on estimating the causality effect during a disaster scenario is published in CIKM 2023.
|
| 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
Currently the research is progressing rather smoothly. With the fast development of the privacy preserving AI, I have made some adjustment to the methods we would like to use to catch up the more latest research trend. More advanced and powerful method is studied and extended under this project, and we have published several papers on this.
|
| 今後の研究の推進方策 |
In the final year, we will continue to study on this direction, with the extension that testing on more different sensing tasks on the smartphone and more advanced data valuation algorithm with the consideration of game theory, which is also the basis of Shapley value in the initial proposal.
|
