IoT devices are sorely underutilized in the medical field, especially within machine learning for medicine, yet they offer unrivaled benefits. IoT devices are low-cost, energy-efficient, small and intelligent devices. In this paper, we propose a distributed federated learning framework for IoT devices, more specifically for IoMT (Internet of Medical Things), using blockchain to allow for a decentralized scheme improving privacy and efficiency over a centralized system; this allows us to move from the cloud-based architectures, that are prevalent, to the edge. The system is designed for three paradigms: 1) Training neural networks on IoT devices to allow for collaborative training of a shared model whilst decoupling the learning from the dataset to ensure privacy. Training is performed in an online manner simultaneously amongst all participants, allowing for the training of actual data that may not have been present in a dataset collected in the traditional way and dynamically adapt the system whilst it is being trained. 2) Training of an IoMT system in a fully private manner such as to mitigate the issue with confidentiality of medical data and to build robust, and potentially bespoke, models where not much, if any, data exists. 3) Distribution of the actual network training, something federated learning itself does not do, to allow hospitals, for example, to utilize their spare computing resources to train network models.

翻译：物联网设备在医疗领域，尤其是医学机器学习中，未得到充分利用，但它们提供了无与伦比的优势。物联网设备是低成本、节能、小型且智能的设备。本文针对物联网设备，更具体地说是针对医疗物联网，提出了一种分布式联邦学习框架，利用区块链实现去中心化方案，从而在隐私和效率方面优于集中式系统；这使我们能够从普遍存在的基于云的架构转向边缘计算。该框架设计围绕三个范式：1）在物联网设备上训练神经网络，以实现共享模型的协作训练，同时将学习过程与数据集解耦以确保隐私。所有参与者以在线方式同时进行训练，这使得能够训练传统采集数据集中可能不存在的数据，并在训练过程中动态调整系统。2）以完全私有方式训练医疗物联网系统，以缓解医疗数据机密性问题，并构建鲁棒且可能定制的模型，即使可用数据很少甚至没有。3）分布式网络训练——这本身并非联邦学习的特性——例如，使医院能够利用其闲置的计算资源来训练网络模型。