Extract And Apply Sparse Connectivity Networks From Multivariate Time Series

Real-world dynamic systems are often monitored with multivariate time series,where each dimension represents a local component measured through a(virtual)sensor.Performing ac-curate diagnostic for a group of dynamic systems while simultaneously taking into account their similarities/distinctions,is a non-trivial task.To this end,This paper develops an effi-cient and adaptive regularization approach,named SAR,to learn sparse connectivity structures in multivariate time series.The learned connectivity networks shed light on the information contained in the systems themselves and hence can serve as highly informative inputs for vari-ous machine learning tasks such as classification.In particular,we focus on high-dimensional and unsupervised(or semi-supervised)learning scenarios and present an efficient parallel learn-ing method,which can identify system-wise connectivity patterns by adaptively constructing a shared,sparsity-inducing regularization template across all systems.Moreover,our approach has the flexibility to incorporate prior information such as must-links and cannot-links for construct-ing more applicable sparse connectivity network.Overall,our approach,named sparse adaptive regularization(SAR),can extract signal-wise connectivity features efficiently and effectively,and result in significant improvements for machine learning tasks.Extensive experiments on real world data demonstrate SAR's superiority over state-of-the-art baselines in terms of accuracy,efficiency,and interpretability.At the same time,in order to transform the numerical information in the matrix into the structural information,this paper proposes three unsupervised learning algorithms to reorder the rows and columns of the matrix,which are SVD imsort,SVD greedy and direct greedy,which are ensembled into an advanced matrix reordering scheme.Experimental results on the gray scale and color images with shuffled rows and columns show that our algorithm can recover the original images.Hence,it proves that the reordering scheme can extract the original structural information of images,which are fundamentally required in structure-aware convolutional neural networks.In addition,we test the algorithm on the sparse connectivity network extracted by SAR model.The experimental results also confirm that our model can uncover the structural information,which can make the sparse connectivity network easily applicable to other tasks.For PPMI data set,the combination of SAR and matrix reordering algorithm can achieve a high accuracy of 94.9% to diagnose Parkinson's disease.This paper has made some contribution to the extraction and application of sparse connectivity network from multivariate time series.