Hybridization of Capsule and LSTM Networks for unsupervised anomaly detection on multivariate data
This addresses the problem of detecting anomalies in multivariate time series without labeled data, which is incremental as it combines existing components (LSTM and Capsule networks) in a novel architecture.
The paper tackles unsupervised anomaly detection on multivariate time series data by introducing a hybrid LSTM-Capsule network in a branched autoencoder architecture, achieving state-of-the-art performance on a benchmark with improved training efficiency and reduced overfitting.
Deep learning techniques have recently shown promise in the field of anomaly detection, providing a flexible and effective method of modelling systems in comparison to traditional statistical modelling and signal processing-based methods. However, there are a few well publicised issues Neural Networks (NN)s face such as generalisation ability, requiring large volumes of labelled data to be able to train effectively and understanding spatial context in data. This paper introduces a novel NN architecture which hybridises the Long-Short-Term-Memory (LSTM) and Capsule Networks into a single network in a branched input Autoencoder architecture for use on multivariate time series data. The proposed method uses an unsupervised learning technique to overcome the issues with finding large volumes of labelled training data. Experimental results show that without hyperparameter optimisation, using Capsules significantly reduces overfitting and improves the training efficiency. Additionally, results also show that the branched input models can learn multivariate data more consistently with or without Capsules in comparison to the non-branched input models. The proposed model architecture was also tested on an open-source benchmark, where it achieved state-of-the-art performance in outlier detection, and overall performs best over the metrics tested in comparison to current state-of-the art methods.