Apparatus and method for determining learning rate using multi-model and gradual batch

摘要

The present invention relates to a technique for determining a learning rate of batch gradient descent used in machine learning, and more particularly, a plurality of models to which different learning rates are applied and a sample size gradually increasing It relates to an apparatus and method for determining a learning rate using a multi-model and progressive batch that can search for an appropriate learning rate using batch. To this end, the apparatus for determining a learning rate using a multi-model and gradual batch according to the present invention includes a batch generator that generates a plurality of batch groups, and applies the plurality of learning rates to the machine learning model to set the machine at each learning rate for the current batch group. A learning unit that learns a learning model and calculates an average loss value for the corresponding learning rate, a learning rate selector that selects a learning rate showing the lowest average loss value by comparing the calculated loss average value, a plurality of based on the selected learning rate and a learning rate changing unit for generating a learning rate.