Initializing LSTM internal states via manifold learning

TL;DR

This paper introduces a manifold learning-based method for initializing LSTM internal states, improving performance and enabling better modeling of partially observed dynamical systems.

Contribution

It proposes a novel approach to initialize LSTM states using data manifold learning, ensuring consistency with observed data and enhancing system identification.

Findings

Improved LSTM performance on a chemical model system

Learned manifold enables transformation of partially observed to fully observed dynamics

Method facilitates alternative nonlinear system identification paths

Abstract

We present an approach, based on learning an intrinsic data manifold, for the initialization of the internal state values of LSTM recurrent neural networks, ensuring consistency with the initial observed input data. Exploiting the generalized synchronization concept, we argue that the converged, "mature" internal states constitute a function on this learned manifold. The dimension of this manifold then dictates the length of observed input time series data required for consistent initialization. We illustrate our approach through a partially observed chemical model system, where initializing the internal LSTM states in this fashion yields visibly improved performance. Finally, we show that learning this data manifold enables the transformation of partially observed dynamics into fully observed ones, facilitating alternative identification paths for nonlinear dynamical systems.