Scholarlink

This study presents Memory enhanced Liquid Time- constant Networks (LTCs) to enhance the temporal and memory dynamics of recurrent models. LTCs, characterized by liquid time constants offer adaptive and stable representations for time-series prediction by modulating hidden states with input- dependent dynamics. Integrating memory features that utilize attractor dynamics to model memory mechanisms explicitly further, augment the system’s representational and computational efficiency. By leveraging the inherent strengths of memory retention and LTCs’ expressivity in capturing continuous time processes, the proposed framework is designed for advanced time-series modeling tasks.

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

Time Series Analysis Using Memory Enhanced Liquid Neural Network

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Abhijeet Kumar,
  • M. Srinivas,
  • R. B. V. Subramanyam

摘要

This study presents Memory enhanced Liquid Time- constant Networks (LTCs) to enhance the temporal and memory dynamics of recurrent models. LTCs, characterized by liquid time constants offer adaptive and stable representations for time-series prediction by modulating hidden states with input- dependent dynamics. Integrating memory features that utilize attractor dynamics to model memory mechanisms explicitly further, augment the system’s representational and computational efficiency. By leveraging the inherent strengths of memory retention and LTCs’ expressivity in capturing continuous time processes, the proposed framework is designed for advanced time-series modeling tasks.