You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
mitsuhiko-nozawa
changed the title
[WIP] One-Bit Over-the-Air Aggregation for Communication-Efficient Federated Edge Learning: Design and Convergence Analysis
One-Bit Over-the-Air Aggregation for Communication-Efficient Federated Edge Learning: Design and Convergence Analysis
Jun 1, 2021
一言でいうと
one-bit broadband digital aggregation (OBDA)で多数のエッジデバイスが同時に通信できるように、コミュニケーションボトルネックを改善
論文リンク
https://arxiv.org/pdf/2001.05713.pdf
著者/所属機関
Guangxu Zhu, Yuqing Du, Deniz G ¨und ¨uz, and Kaibin Huang
Shenzhen Research Institute of Big Data, Shenzhen, China
The University of Hong Kong, Hong Kong
Imperial College of London, London, U.K.
投稿日付(yyyy/MM/dd)
IEEE 2020/11/26
概要
Federated Learningにおいて、同時にパラメータサーバーにアクセスする際に問題となるネットワークボトルネックを、勾配の符号を取る量子化手法と、集約(averaging, voting)された勾配を無線通信時に推定することで改善
新規性・差分
従来の研究はアナログ変調に関するものが多かったが、現在普及しているデジタル変調で利用できるようにした
手法
各エッジデバイスで定期的に計算された勾配の符号を、エッジサーバー(パラメータサーバー)に同時に通信、集約とボーティング後、再びエッジデバイスに通信することを繰り返す
結果
1つのエッジサーバー、100個のエッジデバイスを使ってMNISTとCifar-10を学習
MNISTでは6層ネットワーク、Cifar-10ではResNet18
3つのwireless hostilities下で測定
収束率は異なるがどのhostility下でも収束はしており、デバイス数が増えると精度が向上している(図右)
コメント
The text was updated successfully, but these errors were encountered: