Exploring federated learning on battery-powered devices

   @inproceedings{10.1145/3321408.3323080,
       author = {Xu, Zichen and Li, Li and Zou, Wenting},
       title = {Exploring Federated Learning on Battery-Powered Devices},
       year = {2019},
       isbn = {9781450371582},
       publisher = {Association for Computing Machinery},
       address = {New York, NY, USA},
       url = {https://doi.org/10.1145/3321408.3323080},
       doi = {10.1145/3321408.3323080},
       abstract = {Smartphones generate private data ubiquitously that serves Big data analysis. Without violating privacy issue, analytic companies want to understand and learn features from these data, which fits the nature of federated learning. Federated learning invites multiple participants to train a learning model (e.g., Artificial Neural Network) while guaranteeing the data privacy. However, processing federated learning is heavy on batteries, which is against the ubiquity feature of smartphone. In this paper, we explore the possibility of enabling federated learning on many battery-powered devices. We share our observations on supporting federated learning using battery power, and propose a two-layered strategy to process the learning on batteries with a reasonable tradeoff. The first layer improves the initialization of the federated learning while the second layer explores local energy saving potential. The results show that our work can finish the learning process successfully, consuming 20\% less energy on average, and pays negligible overhead (average 0.1\%) and accuracy loss (2\% at most), as compared to the default setting.},
       booktitle = {Proceedings of the ACM Turing Celebration Conference - China},
       articleno = {6},
       numpages = {6},
       location = {Chengdu, China},
       series = {ACM TURC '19}
   }