Original title: AdaptiveFL: Adaptive Heterogeneous Federated Learning for Resource-Constrained AIoT Systems
Authors: Chentao Jia, Ming Hu, Zekai Chen, Yanxin Yang, Xiaofei Xie, Yang Liu, Mingsong Chen
The article explores improving Federated Learning (FL) for AIoT devices that have different abilities and work in uncertain environments. They introduce AdaptiveFL, a clever approach using a new method called fine-grained width-wise model pruning. This creates diverse models to match the varied capacities of AIoT devices. AdaptiveFL also uses reinforcement learning to pick the best models for each device as they start training, adapting on the go based on their resources. Results from tests show that AdaptiveFL outperforms other methods by up to 16.83% in making predictions, whether the data on devices is similar or different. Essentially, it’s a smarter way for devices to learn together, tailoring the training to each device’s strengths, and significantly improving their collective performance in real-world conditions.
Original article: https://arxiv.org/abs/2311.13166