METHOD FOR ADAPTIVELY CONTROLING PRECISION OF ACTIVATION AND WEIGHT OF ARTIFICIAL NEURAL NETWORK AND APPARATUS THEREOF

摘要

The present invention relates to a method and apparatus for adaptively controlling the precision of an active value and a weight value of an artificial neural network, and a result obtained by expressing the precision with the maximum bit for input data and the result of expressing the precision with N bits for the operation result determining a result value obtained by representing the precision with a value and (Nk) bits; determining a trigger value for controlling the number of bits for indicating the precision by using the respective result values; The present invention relates to a method of controlling the precision of an active value and a weight value of an artificial neural network, comprising the step of controlling the number of bits for indicating the precision by comparing the trigger value with a plurality of different threshold values.