Applying LSTM to Time Series Predictable through Time-Window Approaches

Gers, Felix A.; Eck, Douglas; Schmidhuber, Jürgen

doi:10.1007/3-540-44668-0_93

Cited by 226 publications

(143 citation statements)

References 17 publications

Supporting

Mentioning

137

Contrasting

Unclassified

Order By: Relevance

“…Email: h.jaeger@iu-bremen.de output in trial j, 2 variance of MGS signal), improving the best previous techniques (9)(10)(11)(12)(13)(14)(15), which used training sequences of length 500 to 10,000, by a factor of 700. If the prediction run was continued, deviations typically became visible after about 1300 steps ( Fig.…”

mentioning

confidence: 99%

Harnessing Nonlinearity: Predicting Chaotic Systems and Saving Energy in Wireless Communication

Jaeger

Haas

2004

Science

3,175

2,205

View full text Add to dashboard Cite

We present a method for learning nonlinear systems, echo state networks (ESNs). ESNs employ artificial recurrent neural networks in a way that has recently been proposed independently as a learning mechanism in biological brains. The learning method is computationally efficient and easy to use. On a benchmark task of predicting a chaotic time series, accuracy is improved by a factor of 2400 over previous techniques. The potential for engineering applications is illustrated by equalizing a communication channel, where the signal error rate is improved by two orders of magnitude.

show abstract

mentioning

confidence: 99%

Harnessing Nonlinearity: Predicting Chaotic Systems and Saving Energy in Wireless Communication

Jaeger

Haas

2004

Science

3,175

2,205

View full text Add to dashboard Cite

show abstract

“…Later, this machine learning method was developed to be the perfect fit for regression problems [20].…”

Section: B Svrmentioning

confidence: 99%

“…Our regression model considered the past observations as the input and the future values as the output. The window approach [20] involves the following: ..., , 2 1 denotes the training values, and n is the window size. After different attempts, we chose to fix the window size at 7 for this case study.…”

Section: A Univariate Modelsmentioning

confidence: 99%

See 1 more Smart Citation

E-Commerce Price Forecasting Using LSTM Neural Networks

Bakir¹,

Chniti²,

Zaher³

2018

IJMLC

View full text Add to dashboard Cite

Abstract-In this paper, we provide a robust forecasting model to predict phone prices in European markets using Long Short-Term Memory (LSTM) neural network and Support Vector Regression (SVR). We propose a comparison study of time series forecasting models for these two techniques. LSTM, due to its architecture, is considered as the perfect solution to problems not resolvable by classic Recurrent Neural Networks (RNNs). On the other hand, Support Vector Machines (SVMs) are a very powerful machine learning method for both classification and regression. After studying and comparing several univariate models, SVR and LSTM neural networks appear to be the most accurate ones. In addition, we compared multivariate models for both these techniques. Considering the multivariate approach, by introducing more variables, we obtain better prediction performance. In fact, the SVR model is able to predict the next day price with an root mean squared error (RMSE) of 33.43 euros with the univariate model. However, using multivariate models, LSTM RNN gives the most accurate prediction for the next day's price with an RMSE of 23.640 euros.Index Terms-Time series forecasting, LSTM neural network, support vector regression, e-commerce data, machine learning, deep learning.

show abstract

“…Some works have detected serious limitations [6,7] of RNNs when applied to nonlinear numeric prediction tasks. The findings presented in this paper suggest similar conclusions.…”

mentioning

confidence: 99%

A comparison between recurrent neural architectures for real-time nonlinear prediction of speech signals

Pérez-Ortiz

Calera-Rubio

Forcada

Neural Networks for Signal Processing XI: Proceedings of the 2001 IEEE Signal Processing Society Workshop (IEEE Cat. No.01TH858

View full text Add to dashboard Cite

Abstract. This paper presents a comparative study on the performance of recurrent neural networks trained in real-time to predict the next sample in a speech signal. The comparison is basically done versus linear predictors, and a pipelined recurrent neural network which has been proposed for this task. Results confirm those of previous works where limitations to deal with numeric time series were detected for recurrent neural architectures, specially when using the real-time recurrent learning algorithm. The decoupled extended Kalman filter training algorithm, on the other hand, overcomes partially some of these limitations.

show abstract

Applying LSTM to Time Series Predictable through Time-Window Approaches

Cited by 226 publications

References 17 publications

Harnessing Nonlinearity: Predicting Chaotic Systems and Saving Energy in Wireless Communication

Harnessing Nonlinearity: Predicting Chaotic Systems and Saving Energy in Wireless Communication

E-Commerce Price Forecasting Using LSTM Neural Networks

A comparison between recurrent neural architectures for real-time nonlinear prediction of speech signals

Contact Info

Product

Resources

About