2021/03/01 by Hyun-Suk Lee, Jang-Won Lee, Lee, Hyun-Suk +1 · 1 citation
Computer Science · Engineering · #Privacy-Preserving Technologies in Data #Cooperative Communication and Network Coding #Advanced MIMO Systems Optimization
paper · pdf · doi:10.48550/arxiv.2103.01422
In this paper, we study asynchronous federated learning (FL) in a wireless\ndistributed learning network (WDLN). To allow each edge device to use its local\ndata more efficiently via asynchronous FL, transmission scheduling in the WDLN\nfor asynchronous FL should be carefully determined considering system\nuncertainties, such as time-varying channel and stochastic data arrivals, and\nthe scarce radio resources in the WDLN. To address this, we propose a metric,\ncalled an effectivity score, which represents the amount of learning from\nasynchronous FL. We then formulate an Asynchronous Learning-aware transmission\nScheduling (ALS) problem to maximize the effectivity score and develop three\nALS algorithms, called ALSA-PI, BALSA, and BALSA-PO, to solve it. If the\nstatistical information about the uncertainties is known, the problem can be\noptimally and efficiently solved by ALSA-PI. Even if not, it can be still\noptimally solved by BALSA that learns the uncertainties based on a Bayesian\napproach using the state information reported from devices. BALSA-PO\nsuboptimally solves the problem, but it addresses a more restrictive WDLN in\npractice, where the AP can observe a limited state information compared with\nthe information used in BALSA. We show via simulations that the models trained\nby our ALS algorithms achieve performances close to that by an ideal benchmark\nand outperform those by other state-of-the-art baseline scheduling algorithms\nin terms of model accuracy, training loss, learning speed, and robustness of\nlearning. These results demonstrate that the adaptive scheduling strategy in\nour ALS algorithms is effective to asynchronous FL.\n