WebFL - WebGPU for Federated Learning

Federated Learning (FL) is a machine learning (ML) approach that allows models to be trained collaboratively and distributedly without sharing individual data points. WebGPU, a modern web standard and API, provides low-level, high-performance access to a device's GPU through web browsers. It aims to enable developers to create complex graphical applications and experiences on the web by utilizing modern GPU hardware for efficient graphics rendering. Due to their extensive usage and customized optimizations, current Federated Learning (FL) systems frequently concentrate on certain device types, such as smartphones or Internet of Things (IoT) devices. Nevertheless, FL's adaptability to other device kinds may be limited by this device-centric optimization, necessitating substantial coding and redesign work to modify FL systems for diverse device types. Because developers wind up building different and fragmented codebases for each device, this method violates software engineering principles of code reusability, making maintenance more difficult and impeding the effective deployment of FL. In this proposal, a novel WebGPU based federated learning network for heterogeneous device network is proposed. The proposal aims to deliver a working federated learning system supported by web browsers along with additional features.