Ju n 20 07 Learning to segment temporal sequences by mixture of RNN experts with adaptive variance model

Abstract

This paper proposes a novel method for learning with a “mixture of experts” type model, which can acquire the ability to generate desired sequences by switching experts dynamically. Our method is based on maximum likelihood estimation using gradient descent algorithm similar to conventional methods, though the likelihood function is modified by adding an… (More)

6 Figures and Tables

Topics

  • Presentations referencing similar topics