A method and system of the deep learning system for parallel processing of a plurality of learning models using time-series data divided by frequency components

摘要

The present invention relates to a deep-learning method and a deep-learning system for performing parallel processing of a plurality of learning models by using time-series data divided according to frequency components. According to one aspect of the present invention, a deep-learning method for learning a model to perform at least one of an operation of predicting a result based on time-series data and an operation of classifying the time-series data includes: a first step of determining a frequency per unit time of the time-series data; a second step of segmenting the determined frequency into a plurality of ranges; a third step of dividing the time-series data into data pieces according to each of the ranges; a fourth step of determining whether an applied learning type is on-line learning or off-line learning; a fifth step of using the divided data pieces as an input to the model according to the determined learning type; and a sixth step of performing, by the model, deep-learning based on the input data.