On the Initialization of Long Short-Term Memory Networks

TL;DR

This paper introduces a new weight initialization method for LSTM networks that improves training stability and convergence, outperforming existing techniques in various time series and disease modeling tasks.

Contribution

A robust initialization approach based on normalized random weights that maintains variance, enhancing LSTM training stability and performance.

Findings

Outperforms state-of-the-art initialization methods

Improves training convergence speed

Enhances generalization in time series tasks

Abstract

Weight initialization is important for faster convergence and stability of deep neural networks training. In this paper, a robust initialization method is developed to address the training instability in long short-term memory (LSTM) networks. It is based on a normalized random initialization of the network weights that aims at preserving the variance of the network input and output in the same range. The method is applied to standard LSTMs for univariate time series regression and to LSTMs robust to missing values for multivariate disease progression modeling. The results show that in all cases, the proposed initialization method outperforms the state-of-the-art initialization techniques in terms of training convergence and generalization performance of the obtained solution.