| Project/Area Number |
23K11080
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 60060:Information network-related
|
|---|
| Research Institution | Ibaraki University |
|---|
Principal Investigator |
王 瀟岩 茨城大学, 理工学研究科(工学野), 准教授 (10725667)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
梅比良 正弘 南山大学, 理工学部, 教授 (00436239)
|
|---|
| Project Period (FY) |
2023-04-01 – 2026-03-31
|
| Project Status |
Granted (Fiscal Year 2023)
|
| Budget Amount *help |
¥4,680,000 (Direct Cost: ¥3,600,000、Indirect Cost: ¥1,080,000)
Fiscal Year 2025: ¥1,040,000 (Direct Cost: ¥800,000、Indirect Cost: ¥240,000)
Fiscal Year 2024: ¥1,950,000 (Direct Cost: ¥1,500,000、Indirect Cost: ¥450,000)
Fiscal Year 2023: ¥1,690,000 (Direct Cost: ¥1,300,000、Indirect Cost: ¥390,000)
|
|---|
| Keywords | split learning / in-network learning / resource constraint |
|---|
| Outline of Research at the Start |
The booming mobile deep learning applications that enable personalized experiences are challenging the current computing and communication network architectures. The state-of-the-art federated learning (FL) and split learning (SL) based solutions, however, are constrained by computational resources and communication resources respectively. To this end, in this research, we propose an efficient split in-network learning approach for resource constrained wireless networks, which amalgamates the FL and SL techniques and eliminates their inherent drawbacks.
|
| Outline of Annual Research Achievements |
In FY2023, we have realized the efficient split in-network learning approach with fixed model and network settings. Specifically, we focused on the realization of BS-side model virtualization and UE-BS model’s gradient aggregation, under a given model partition and UE set. We validated the proposed approach’s convergence speed and model accuracy by simulations on PyTorch. Different from most of the previous studies that use MNIST dataset (i.e., handwritten digits from 0 to 9), we took into consideration the real mobile deep learning applications, and thus used an aerial view human action detection dataset for evaluation.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
We consider that the research progresses smoothly. We have published 1 journal papers, 2 internal conference papers and multiple domestic conference papers in FY2023 under the support of this funding.
|
| Strategy for Future Research Activity |
In the following year, we will consider to refine the approach and clarify the performance tradeoff curve under various resource budget. Furthermore, we will consider to implement the proposed approach on testbed.
|
