2023 Fiscal Year Research-status Report
Over-the-Air連合学習のためのデバイススケジューリングに関する研究
| Project/Area Number |
23KJ1005
|
|---|
| Research Institution | National Institute of Informatics |
|---|
Principal Investigator |
CHEN CAIJUAN 国立情報学研究所, アーキテクチャ科学研究系, 特別研究員(PD)
|
|---|
| Project Period (FY) |
2023-04-25 – 2025-03-31
|
| Keywords | Convergence analysis / Client selection / Energy management |
|---|
| Outline of Annual Research Achievements |
This fiscal year's research concentrated on scheduling devices for over-the-air federated learning to improve the convergence speed of the system, addressing the challenges posed by weak channels and limited energy resources. The study involved a comprehensive literature review, model development, algorithm design, and simulation experiments. The findings demonstrate that optimizing the system model’s convergence analysis, device scheduling, and energy management, considering factors like transmission power, wireless channels, and energy constraints, can significantly enhance the system performance for over-the-air federated learning.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
Due to the availability of equipment in the laboratory for conducting simulations, such as GPUs and CPUs, and with the effective guidance of my supervisor, combined with regular and productive communication with both the supervisor and other collaborators, the overall research progress is proceeding as planned.
|
| Strategy for Future Research Activity |
For the future research plan, we plan to optimize two-tier device scheduling under weak channel conditions based on previous studies for over-the-air federated learning. Furthermore, we will explore the configuration and optimization of reconfigurable intelligent surfaces to enhance the system performance of over-the-air federated learning with weak channels. In our prior research, the primary challenges involved establishing robust optimization models and developing effective optimization methods. To address these challenges and ensure the smooth progress of our future work, we will need to delve into additional relevant literature and acquire advanced optimization techniques.
|
| Causes of Carryover |
In the following fiscal year, we have outlined several key expenditures to support our research development. First, we plan to attend two international conferences, with an estimated total cost of 700,000 yen, covering conference registration fees, travel fees, and accommodation fees. Additionally, we intend to participate in two domestic conferences, with an estimated total cost of 300,000 yen, including travel fees and accommodation fees. To facilitate our research, we plan to purchase Raspberry Pi devices and GPU cards, with a projected expense of 300,000 yen. We also foresee spending 300,000 yen on publication fees for two international journals to disseminate our research findings. Moreover, to ensure the smooth progress of our research activities, we need to purchase specialized software, including MATLAB and Adobe, with an anticipated cost of 200,000 yen.
|
