Search:
Match:
1 results

Analysis

This paper addresses the challenge of enabling efficient federated learning in space data centers, which are bandwidth and energy-constrained. The authors propose OptiVote, a novel non-coherent free-space optical (FSO) AirComp framework that overcomes the limitations of traditional coherent AirComp by eliminating the need for precise phase synchronization. This is a significant contribution because it makes federated learning more practical in the challenging environment of space.
Reference

OptiVote integrates sign stochastic gradient descent (signSGD) with a majority-vote (MV) aggregation principle and pulse-position modulation (PPM), where each satellite conveys local gradient signs by activating orthogonal PPM time slots.