| CPC G06N 3/0985 (2023.01) [G06F 11/3447 (2013.01); G06N 20/00 (2019.01)] | 20 Claims |
|
1. A method, comprising:
determining an initial learning rate combination for a deep learning model in a processor-based machine learning system, wherein the initial learning rate combination comprises a plurality of learning rates, each learning rate being determined for one of a plurality of layers of the deep learning model, and the plurality of learning rates comprising static learning rates and dynamic learning rates;
generating a coded representation of the initial learning rate combination in the processor-based machine learning system, wherein the coded representation comprises a plurality of entries for respective ones of the layers of the deep learning model, with each of the entries being configured to include a value denoting a particular selected one of the static learning rates and the dynamic learning rates for its corresponding one of the layers;
adjusting the initial learning rate combination in the processor-based machine learning system to obtain a target learning rate combination, by application of an annealing algorithm of the processor-based machine learning system to the coded representation;
training the deep learning model in the processor-based machine learning system utilizing the target learning rate combination; and
performing a recognition task in the processor-based machine learning system utilizing the trained deep learning model;
wherein an accuracy rate achieved by the processor-based machine learning system for the recognition task when the target learning rate combination is used to train the deep learning model is higher than or equal to a first threshold accuracy rate.
|