Advanced deep graph neural networks for explainable anomaly detection study
| Project/Area Number |
22K17961
|
|---|
| Research Category |
Grant-in-Aid for Early-Career Scientists
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Basic Section 61030:Intelligent informatics-related
|
|---|
| Research Institution | National Institute of Informatics (2023)
National Institute of Advanced Industrial Science and Technology (2022) |
|---|
Principal Investigator |
Ouyang Tinghui 国立情報学研究所, 情報社会相関研究系, 特任研究員 (80870849)
|
|---|
| Project Period (FY) |
2022-04-01 – 2025-03-31
|
| Project Status |
Granted (Fiscal Year 2023)
|
| Budget Amount *help |
¥3,770,000 (Direct Cost: ¥2,900,000、Indirect Cost: ¥870,000)
Fiscal Year 2024: ¥1,040,000 (Direct Cost: ¥800,000、Indirect Cost: ¥240,000)
Fiscal Year 2023: ¥1,170,000 (Direct Cost: ¥900,000、Indirect Cost: ¥270,000)
Fiscal Year 2022: ¥1,560,000 (Direct Cost: ¥1,200,000、Indirect Cost: ¥360,000)
|
|---|
| Keywords | Anomaly detection / Out-of-distribution / data structure / explaination / granular computing / data mining / anomaly data detection / information graph / structure description / explanation / graph neural network / anomaly detection / explainability / big data analysis |
|---|
| Outline of Research at the Start |
Aim at challenges of anomaly detection study related to big data and deep learning, an advanced graph neural network model is proposed. This research granulates big data in modeling to reduce computation cost, and leverage graph structure to provide good explainability for DL-based AD model.
|
| Outline of Annual Research Achievements |
In terms of explanation of anomaly detection, we proposed granule-based anomalous data descriptor and detector for explanation. Moreover, we developed a structure matrix which is useful to realize data structure contraction and helpful to explain that anomaly data usually have large distance in the process of data structure contraction. In terms of anomaly detection applications, we apply the proposed granular AD detector to detect the out-of-distribution data in image and textual data. Moreover, a data quality assurance issue is discussed based on GPT-based sentiment analysis application.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
According to the proposal, our current progress is going as expected without delays. There are totally four conference papers and one journal submitted, among which two have been published in this fiscal year. In the next step, we will develop more anomaly detection algorithms and applications based on granular structure and granular information graph.
|
| Strategy for Future Research Activity |
According to the plan in the proposal, we will do more research on developing advanced anomaly data detection algorithms and applications. One is to construct granular information graph, to develop GNN for anomaly detection, and to provide explanation. Then, we will continue solving various practical problems and applications related to anomaly data detection, and provide explanation for the AD process.
|
Report
(2 results)
Research Products
(8 results)
