NIAA: Neuroplasticity-inspired adaptive aggregation method for federated learning

Abstract

Federated Learning (FL) has emerged as a promising solution for distributed machine learning by enabling decentralized model training across edge devices and preserving the privacy of data. However, the effectiveness of FL heavily depends on the aggregation strategy used to integrate client updates, especially in the presence of non-IID data, unreliable participation, and noisy local training. Traditional approaches such as Federated Averaging (FedAvg) and recent meta-heuristicbased strategies often fail to incorporate client behavioral patterns over time, resulting in suboptimal convergence and fairness. In this paper, we propose NIAA, a Neuroplasticity-Inspired Adaptive Aggregation method, that dynamically adjusts client aggregation weights by modeling synaptic strength as a memory-driven function of effectiveness and update stability. Inspired by biological learning mechanisms, NIAA reinforces contributions from clients that consistently improve the global model while attenuating the influence of unstable or erratic participants. Experimental evaluations on the MNIST dataset under both IID and non-IID settings demonstrate that NIAA significantly outperforms state-of-the-art baselines in terms of accuracy, loss reduction, and robustness to data heterogeneity, establishing a biologically grounded paradigm for adaptive FL aggregation.