Privacy-Preserving and Secure Divide-and-Conquer Learning

Abstract

The computation need of neural networks has out-paced the capabilities of many individual users whose computers, mobile devices, and other devices are relatively limited in computation power. To solve this problem, currently users need to offload the model training task to the cloud that has many computing resources. On the other hand, many devices on the edge have idling CPU cycles not used. Inspired by the successes of crowdsourcing and decentralized computing platforms such as blockchain and Web3, we propose to outsource an individual's neural network training task to edge devices, such that individuals can train their own neural network models without relying on the centralized cloud. Specifically, we design a divide-and-conquer learning framework in the edge computing environment. A user can divide the training computation of its neural network into neuron-sized computation tasks and distribute them to devices in the edge based on their available resources. The results will be returned to the user and aggregated in an iterative process to obtain the final neural network model. To protect the privacy of the user's data and model, shuffling is done to both the data and the neural network model before the computation task is distributed to edge nodes. Security against misbehaving edge nodes can also be provisioned by redundancy in task assignment.