In order to fit today’s neural networks onto hardware, some practitioners utilize some type of weight-pruning method to compress the size of the model and reduce the size and computational cost of running it. Today’s weight-pruning methods can reach up to 50× data compression without major accuracy loss This is on top of ephemeral activation […] The post Can a Sparse-AI Hardware Architecture for Data Centers Work? appeared first on HPCwire .
Source: HPCwire — read the full report at the original publisher.