IntroductionToday, the development of mobile devices and the Internet of ings (IoT) with many applications has re-cently brought benets to society, from smart cities and intelligent transportation to smart cities and personalized health care support systems. Articial intelligence (AI) [8] has support data from many dierent information sources that can be modeled through training processes. erefore, the trained model is used in future forecasting tasks. Besides, the development of telecommunications technol-ogy has promoted the network infrastructure, especially in the 5G / 6G era. Today’s intelligent systems have become more data-intensive and high-speed, with low latency and high communication capacity. In 2016, the Japanese gov-ernment proposed building a Japanese Society 5.0*1, a super-smart society. It takes advantage of the most outstanding achievements of the industrial revolution 4.0 (i.e., articial intelligence, robotics, internet of things, big data) to solve complex social problems and bring a whole life to people in the future. Japanese Society 5.0 Beyond 5G / 6G [1] is an indispensable component supporting articial intelli-gence, big data, and other ancillary technologies.Although, modern deep learning models and big data analytics based on AI models oen require enormous computing power and data consumption. However, thanks to the development of telecommunications technology, 1従来の機械学習パイプラインでは、モデルを学習するために単一のコーパスに収集されたデータを必要としていた。機械学習手法の一種である連合学習は、複数の分散型エッジデバイスやデータ収集にわたってモデルを学習する仕組みを提供する。エッジと中央サーバー間の通信要件や異種データの分布とインフラストラクチャは、実世界のアプリケーションにおける連合学習の課題となっている。本稿では、連合学習の包括的な定義を提供・分類し、本質的な課題について議論する。Non-IID分布の問題や計算限界に対処するため、グローバル側での連合学習連携アルゴリズムとエッジAI適応アプローチを導入した。提案した技術の環境及びスマートホームサービスなどへの応用についても述べる。また、Beyond 5G / 6G適応に向けた新しい研究の方向性について述べる。Conventional machine learning pipelines require data collected into a single corpus to train models. Federated learning (FL), a type of machine learning technique, provides a mechanism to train models across multiple decentralized edge devices or data collection. The requirement of communication among edges and central server and the heterogeneous data distribution and infra-structure challenges federated learning in real world application. In this paper, we provide a com-prehensive definition of federated learning, categorize it, and discuss essential challenges. To tackle non-IID distribution issues and computational limitations at edges, we introduced federated learning collaboration algorithms on the global side and Edge AI adaptation approach. The proposed techniques have been applied to environmental and smart home services, etc. We also address new research directions in the context of Beyond 5G / 6G adaptation.4-4 パブリック・プライベートデータを活用した予測モデリングを実現する連合学習によるエッジAI4-4Edge AI with Federated Learning for Public-private Cooperative Predictive Modelingグエン ド ヴァン チャン アイン クアNGUYEN Do-Van and TRAN Anh Khoa＊1https://www8.cao.go.jp/cstp/english/society5_0/index.html1774 スマートデータ利活用基盤技術
元のページ ../index.html#183