Time2Graph: Revisiting Time Series Modeling with Dynamic Shapelets

Cheng, Ziqiang; Yang, Yang; Wang, Wei; Hu, Wenjie; Zhuang, Yueting; Song, Guojie

doi:10.1609/aaai.v34i04.5769

Cited by 48 publications

(23 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To satisfy this need, we propose a twophase approach to create the MAFNs from EEG readings. Furthermore, according to different calculation indexes of each stage, three kinds of are constructed, based on dynamic time warping [34] (MAFN-dtw), symbolic mutual information [35,36] (MAFN-smi), and hub depressed index [37] (MAFN-HDI) respectively.…”

Section: Construction Of Mafnsmentioning

confidence: 99%

“…First, to reveal temporal regularities an idea from network science (see [30,31]) is adopted for building a network from a single time series. As illustrated in Figure 1, we use a sliding window to split all EEG time series (Figure 1B) into m subsequences (Figure 1C), and then identify from these m subsequences the representative sub-sequences (RSs) through the idea similar to clustering: 1) Calculate the similarity (measured by dynamic time warping (DTW) [34] or symbolic mutual information (SMI) [35,36]) between each pair of sub-sequences, and for each sub-sequence we select its k most similar (other) sub-sequences; 2) In the mk selected sub-sequences, some are repeated. Hence we pick the top k ones who occur most frequently.…”

Section: Construction Of Mafnsmentioning

confidence: 99%

See 1 more Smart Citation

Multilayer-Aggregation Functional Network for Identifying Brain Fatigue and Diseases

Cui

Qi²,

Sun³

et al. 2022

Front. Phys.

View full text Add to dashboard Cite

Recent years have witnessed increasing interest of applying network science methodologies to analyze brain activity data. Owing to the noninvasiveness, low cost and high sampling rate, electroencephalogram (EEG) recordings have been widely used as a proxy for probing the internal states of human brains. Previous correlation-based functional networks (CFN) mainly focused on the covariance or coherence between readings from electrodes attached to different regions, largely overlooking local temporal properties of these electrical activities. Here, we propose a method to construct multilayer-aggregation functional network (MAFN) which is able to capture both temporal and topological characteristics from EEG data. We extract features from these MAFNs and incorporate them into each of 12 classification algorithms, aiming to detect mental fatigue and two brain diseases, schizophrenia and epilepsy. The results demonstrate that MAFNs consistently outperform CFN and dynamic version of CFN. In comparison to functional networks based on weighted phase lag index (wPLI), MAFNs also achieve higher or comparable accuracy in most classifiers. Moreover, the nodal features of MAFNs allow us to identify the important positions of EEG electrodes for different brain states or diseases. These findings together offer not only a framework for classifying normal and abnormal brain activities but also a general method for constructing more informative functional networks from multiple time series data.

show abstract

Section: Construction Of Mafnsmentioning

confidence: 99%

Section: Construction Of Mafnsmentioning

confidence: 99%

Multilayer-Aggregation Functional Network for Identifying Brain Fatigue and Diseases

Cui

Qi²,

Sun³

et al. 2022

Front. Phys.

View full text Add to dashboard Cite

show abstract

“…These methods ignore the relationships among the time series, thus making it difficult to understand the causes of outliers. In a previous study [ 9 ], time-aware shapelets were extracted to construct an evolution graph to detect time series outliers. The approach was applied on a single signal, and outliers were detected only by comparing the signal with itself.…”

Section: Introductionmentioning

confidence: 99%

Dynamic graph embedding for outlier detection on multiple meteorological time series

Jung

2021

PLoS ONE

View full text Add to dashboard Cite

Existing dynamic graph embedding-based outlier detection methods mainly focus on the evolution of graphs and ignore the similarities among them. To overcome this limitation for the effective detection of abnormal climatic events from meteorological time series, we proposed a dynamic graph embedding model based on graph proximity, called DynGPE. Climatic events are represented as a graph where each vertex indicates meteorological data and each edge indicates a spurious relationship between two meteorological time series that are not causally related. The graph proximity is described as the distance between two graphs. DynGPE can cluster similar climatic events in the embedding space. Abnormal climatic events are distant from most of the other events and can be detected using outlier detection methods. We conducted experiments by applying three outlier detection methods (i.e., isolation forest, local outlier factor, and box plot) to real meteorological data. The results showed that DynGPE achieves better results than the baseline by 44.3% on average in terms of the F-measure. Isolation forest provides the best performance and stability. It achieved higher results than the local outlier factor and box plot methods, namely, by 15.4% and 78.9% on average, respectively.

show abstract

“…Recently, self-supervised learning has attracted more and more attention in computer vision by designing different pretext tasks on image data such as solving jigsaw puzzles (Noroozi & Favaro, 2016), inpainting (Pathak et al, 2016), rotation prediction (Gidaris et al, 2018), and contrastive learning of visual representations (Chen et al, 2020), and on video data such as object tracking Here, an example of 3-scale temporal relations including short-term, middle-term and long-term temporal relation is given for illustration. (Wang & Gupta, 2015), and pace prediction (Wang et al, 2020). Although some video-based approaches attempt to capture temporal information in the designed pretext task, time series is far different structural data compared with video.…”

Section: Introductionmentioning

confidence: 99%

“…Ensemble-based methods aims at combining multiple classifiers for higher classification performance. More recently, deep learning based methods (Karim et al, 2017;Ma et al, 2019;Cheng et al, 2020) conduct classification by cascading the feature extractor and classifier based on MLP, RNN, and CNN in an end-to-end manner. Our approach focuses instead on self-supervised representation learning of time series on unlabeled data, exploiting inter-sample relation and intra-temporal relation of time series to guide the generation of useful feature.…”

Section: Introductionmentioning

confidence: 99%

Self-Supervised Time Series Representation Learning by Inter-Intra Relational Reasoning

Fan,

Zhang,

Gao

2020

Preprint

View full text Add to dashboard Cite

Self-supervised learning achieves superior performance in many domains by extracting useful representations from the unlabeled data. However, most of traditional self-supervised methods mainly focus on exploring the inter-sample structure while less efforts have been concentrated on the underlying intra-temporal structure, which is important for time series data. In this paper, we present Self-Time: a general Self-supervised Time series representation learning framework, by exploring the inter-sample relation and intra-temporal relation of time series to learn the underlying structure feature on the unlabeled time series. Specifically, we first generate the inter-sample relation by sampling positive and negative samples of a given anchor sample, and intra-temporal relation by sampling time pieces from this anchor. Then, based on the sampled relation, a shared feature extraction backbone combined with two separate relation reasoning heads are employed to quantify the relationships of the sample pairs for inter-sample relation reasoning, and the relationships of the time piece pairs for intra-temporal relation reasoning, respectively. Finally, the useful representations of time series are extracted from the backbone under the supervision of relation reasoning heads. Experimental results on multiple real-world time series datasets for time series classification task demonstrate the effectiveness of the proposed method. Code and data are publicly available at https://haoyfan.github.io/.

show abstract

Time2Graph: Revisiting Time Series Modeling with Dynamic Shapelets

Cited by 48 publications

References 34 publications

Multilayer-Aggregation Functional Network for Identifying Brain Fatigue and Diseases

Multilayer-Aggregation Functional Network for Identifying Brain Fatigue and Diseases

Dynamic graph embedding for outlier detection on multiple meteorological time series

Self-Supervised Time Series Representation Learning by Inter-Intra Relational Reasoning

Contact Info

Product

Resources

About