The present application belongs to the technical field of distributed
machine learning and edge intelligence, and discloses a self-adaptive quantization decentralized learning method and
system for heterogeneous edge devices. The present application is directed to a plurality of edge devices with heterogeneous computing resources, heterogeneous storage capabilities and time-varying communication links. Under the condition of no central
server participation, through training parameter and communication relationship initialization, training quantization scale self-adaptive determination, low-precision local training, communication quantization scale self-adaptive determination, quantization model
information exchange and neighbor aggregation update, decentralized collaborative learning is completed. The present application can simultaneously alleviate the resource limitation problem of heterogeneous edge devices in the training stage and the communication stage, ensure the implementability of decentralized collaborative learning, improve
resource utilization efficiency and improve model convergence stability.