Scholarlink

The wider application of gated models for deep learning requires the design of more effective training algorithms from the point of view of the speed of convergence. The work is devoted to developing a new method of training for the LSTM model. The approach is based on getting Lyapunov stability conditions. The research of qualitative behavior of the model is conducted from the viewpoint of dynamical systems. The method of stable learning was compared with traditional backpropagation through time using one computational example.

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

On Method of Training of LSTM Neural Network Based on Stability Conditions

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Jacek Kafel-Kania,
  • Vasyl Martsenyuk,
  • Georgii Dimitrov,
  • Marcin Bernas

摘要

The wider application of gated models for deep learning requires the design of more effective training algorithms from the point of view of the speed of convergence. The work is devoted to developing a new method of training for the LSTM model. The approach is based on getting Lyapunov stability conditions. The research of qualitative behavior of the model is conducted from the viewpoint of dynamical systems. The method of stable learning was compared with traditional backpropagation through time using one computational example.