Tsukada Lab. Computer Network Group

Wireless Ad-Hoc Federated Learning for Cooperative Map Creation and Localization Models

Yusuke Sugizaki, Hideya Ochiai, Muhammad Asad, Manabu Tsukada, Hiroshi Esaki

9th IEEE World Forum on Internet of Things (WF-IoT 2023), Aveiro, Portugal, 2023年10月12日

3行要約

  • Wi-Fiによる位置推定には、事前に電波強度を計測する必要があることが多い。
  • 無線アドホック連合学習を使い、普段の移動中に協力して地図と位置推定モデルを作る。
  • 実環境で計測したWi-Fi信号強度を用い、複数の分類モデルを組み合わせて位置を推定した。

アブストラクト(原文)

Although Wi-Fi signals have been used for localization, many existing methods require gathering Wi-Fi information about the area in advance. This study proposed a novel system in which wireless ad-hoc federated learning is used to learn localization models and create maps cooperatively during regular movement. In this system, a combination of classification models is used to perform localization from Wi-Fi signal strength measured as received signal strength indicator (RSSI). In this study, RSSI data in a real-world Wi-Fi environment were collected to train and test localization models. The proposed method achieved localization accuracy between 91.30% and 96.11 %, which demonstrated the ability of the method to train localization models collaboratively.

引用方法

Y. Sugizaki, H. Ochiai, M. Asad, M. Tsukada, and H. Esaki, “Wireless Ad-Hoc Federated Learning for Cooperative Map Creation and Localization Models,” in 9th IEEE World Forum on Internet of Things (WF-IoT 2023), Aveiro, Portugal, 2023, doi: 10.1109/WF-IoT58464.2023.10539517.

BibTeX

@inproceedings{Sugizaki2023,
  title         = {Wireless Ad-Hoc Federated Learning for Cooperative Map Creation and Localization Models},
  author        = {Yusuke Sugizaki and Hideya Ochiai and Muhammad Asad and Manabu Tsukada and Hiroshi Esaki},
  doi           = {10.1109/WF-IoT58464.2023.10539517},
  year          = {2023},
  date          = {2023-10-12},
  urldate       = {2023-10-12},
  booktitle     = {9th IEEE World Forum on Internet of Things (WF-IoT 2023)},
  address       = {Aveiro, Portugal},
  abstract      = {Although Wi-Fi signals have been used for localization, many existing methods require gathering Wi-Fi information about the area in advance. This study proposed a novel system in which wireless ad-hoc federated learning is used to learn localization models and create maps cooperatively during regular movement. In this system, a combination of classification models is used to perform localization from Wi-Fi signal strength measured as received signal strength indicator (RSSI). In this study, RSSI data in a real-world Wi-Fi environment were collected to train and test localization models. The proposed method achieved localization accuracy between 91.30% and 96.11 %, which demonstrated the ability of the method to train localization models collaboratively.},
  keywords      = {},
  pubstate      = {published},
  tppubtype     = {inproceedings}
}