Abstract:Echo state networks (ESNs) are randomly connected recurrent neural networks (RNNs) that can be used as a temporal kernel for modeling time series data, and have been successfully applied on time series prediction tasks. Recently, ESNs have been applied to time series classification (TSC) tasks. However, previous ESN-based classifiers involve either training the model by predicting the next item of a sequence, or predicting the class label at each time step. The former is essentially a predictive model adapted from time series prediction work, rather than a model designed specifically for the classification task. The latter approach only considers local patterns at each time step and then averages over the classifications. Hence, rather than selecting the most discriminating sections of the time series, this approach will incorporate non-discriminative information into the classification, reducing accuracy. In this paper, we propose a novel end-to-end framework called the Echo Memory Network (EMN) in which the time series dynamics and multi-scale discriminative features are efficiently learned from an unrolled echo memory using multi-scale convolution and max-over-time pooling. First, the time series data are projected into the high dimensional nonlinear space of the reservoir and the echo states are collected into the echo memory matrix, followed by a single multi-scale convolutional layer to extract multi-scale features from the echo memory matrix. Max-over-time pooling is used to maintain temporal invariance and select the most important local patterns. Finally, a fully-connected hidden layer feeds into a softmax layer for classification. This architecture is applied to both time series classification and human action recognition datasets. For the human action recognition datasets, we divide the action data into five different components of the human body, and propose two spatial information fusion strategies to integrate the spatial information over them. With one training-free recurrent layer and only one layer of convolution, the EMN is a very efficient end-to-end model, and ranks first in overall classification ability on 55 TSC benchmark datasets and four 3D skeleton-based human action recognition tasks.

Echo State Networks for Static Classification with Corresponding Clusters

Analog Circuit Fault Diagnosis with Echo State Networks Based on Corresponding Clusters

Random Subspace and Multiple Reservoirs Based Classifier

An Image Classification Method Based on Echo State Network

Functional Echo State Network for Time Series Classification.

Exploring deep echo state networks for image classification: a multi-reservoir approach

Regular Echo State Networks

Multiple Clusters Echo State Network for Chaotic Time Series Prediction

Sequence Prediction and Classification of Echo State Networks

Time Series Classification with Echo Memory Networks

Time Series Classification Based on Forward Echo State Convolution Network

Performance Evaluation of New Echo State Networks Based on Complex Network

Determining Reservoir Topologies from Short-Term Memory in Echo State Networks.

Echo Memory-Augmented Network for Time Series Classification.

Long-Short Term Echo State Network for Time Series Prediction

Fuzzy Echo State Networks

The Architecture of Dynamic Reservoir in the Echo State Network

Modeling Neural Plasticity in Echo State Networks for Classification and Regression

Design of Sparse Bayesian Echo State Network for Time Series Prediction

Design of Echo State Network with Multi-Objective Optimization Algorithm and L1 Regularization

A priori data-driven multi-clustered reservoir generation algorithm for echo state network.