Predictive GAN-powered Multi-Objective Optimization for Hybrid Federated Split Learning

As an edge intelligence algorithm for multi-device collaborative training, federated learning (FL) can protect data privacy but increase the computing load of wireless devices. In contrast, split (SL) reduce devices by model splitting and assignment. To take advantage FL SL, we propose a hybrid (HFSL) framework networks in this paper, which combines multi-worker training flexible SL. computational idleness splitting, design parallel scheme without label sharing conduct theoretical analysis impact delayed gradient on convergence. Aiming to obtain trade-off between time energy consumption, joint optimization problem decisions, bandwidth, resources as multi-objective problem. such, predictive generative adversarial network (GAN)-powered Pareto front problem, utilizes discriminator guide generator predict promising solutions. Experimental results demonstrate that proposed outperforms considered baselines finding optimal solutions, solutions obtained from HFSL dominate solution FL.