Scholarlink

This work we propose the adaptation the Momentum parameter, which contributes to the step of updating the parameters of the previous iteration of the stochastic gradient descent used Momentum, which value is scalar between 0 and 1. The value of 0 tells us that there is no contribution from the previous step and 1 indicates that the contribution is maximal. It is proposed to find the best value for the contribution towards the next step since this value influences notably in the recognition of the images. It was demonstrated that the optimization of the parameters with a convolutional neuronal network utilizing fuzzy logic helps obtain the maximize the recognition of the images since this value of the Momentum is not fixed. Therefore, it adapts to the previous step making the downward gradient to be reduced notably and consequently the recognition of the images not only increases but it is also faster since better results are obtained with a decrease in the number of training epochs.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Momentum Adaptation in Convolutional Neural Networks Using a Fuzzy Gravitational Algorithm

  • 章节

  • 发表日期：

    2025年

查看全文
  • Yutzil Poma,
  • Patricia Melin

摘要

This work we propose the adaptation the Momentum parameter, which contributes to the step of updating the parameters of the previous iteration of the stochastic gradient descent used Momentum, which value is scalar between 0 and 1. The value of 0 tells us that there is no contribution from the previous step and 1 indicates that the contribution is maximal. It is proposed to find the best value for the contribution towards the next step since this value influences notably in the recognition of the images. It was demonstrated that the optimization of the parameters with a convolutional neuronal network utilizing fuzzy logic helps obtain the maximize the recognition of the images since this value of the Momentum is not fixed. Therefore, it adapts to the previous step making the downward gradient to be reduced notably and consequently the recognition of the images not only increases but it is also faster since better results are obtained with a decrease in the number of training epochs.