FractalWorkspace

It is a distributed computing framework for federated learning across mobile edge devices, with a central server orchestrating tasks and Android nodes doing on-device training or inference. ^[1a][1b] In federated learning, raw training data is not sent to the central server; the server dispatches task descriptors to Android nodes and collects model checkpoint updates. ^[2] It can also split a large foundation model into memory-bounded partitions so Android nodes run different layers and pass compressed activation tensors across a local mesh. ^[3]