NEURAL NETWORK PARAMETER PROCESSING METHOD AND DEVICE IN DISTRIBUTED CLOUD ENVIRONMENT

摘要

The present disclosure provides the steps of: obtaining checkpoint data including a stored set of weight vectors; dividing the checkpoint data into a plurality of checkpoint data blocks; obtaining learning result data; Dispersion of a device comprising the steps of dividing into result data blocks, determining an amount of change between the plurality of checkpoint data blocks and the plurality of learning result data blocks, and updating the checkpoint data based on the determination result Disclosed is a method for processing neural network parameters in a cloud environment.