Article
Physics, Multidisciplinary
Sina Molavipour, Hamid Ghourchian, German Bassi, Mikael Skoglund
Summary: Novel approaches using neural networks to estimate information measures have gained recognition in recent years. The study demonstrates the consistency of these estimators under certain conditions and extends the proof of convergence for complex measures. The effectiveness of the approach is demonstrated through simulations of directed information estimation.
Article
Computer Science, Information Systems
Van Long Ho, Nguyen Ho, Torben Bach Pedersen
Summary: This paper presents a method for frequent temporal pattern mining from time series data, which is efficient and has low memory consumption. It uses an end-to-end process and an efficient algorithm that outperforms baselines in runtime and memory consumption. Additionally, an approximate version of the algorithm provides even faster results while maintaining high accuracy.
PROCEEDINGS OF THE VLDB ENDOWMENT
(2021)
Article
Computer Science, Information Systems
Shuye Pan, Peng Wang, Chen Wang, Wei Wang, Jianmin Wang
Summary: This paper proposes an efficient approach for searching correlated window pairs in two long time series. The approach quickly finds high-quality candidate pairs using two strategies and refines them through a nested one-dimensional search. The empirical study demonstrates the efficiency and effectiveness of the approach on both synthetic and real-world datasets.
PROCEEDINGS OF THE VLDB ENDOWMENT
(2022)
Article
Physics, Multidisciplinary
Muhammad Sheraz, Silvia Dedu, Vasile Preda
Summary: This study empirically examines long memory and bi-directional information flow between estimated volatilities of five cryptocurrencies, confirming their long-run dependence and non-linear behavior. OHLC volatility estimators show significant performance in quantifying information flow, providing an additional choice for comparison with other volatility estimators.
Article
Computer Science, Artificial Intelligence
Dailys M. A. Reyes, Renata M. C. R. de Souza, Adriano L. de Oliveira
Summary: This paper introduces a new three-stage approach for time series forecasting based on quartile data, combining aggregation, prototype selection, and forecasting in the Symbolic Data Analysis (SDA) framework. By summarizing time series and representing classes of entities, a more representative class dataset is obtained through a prototype selection process based on mutual information. Each class is described by a list of three continuous values, known as quartile symbolic data, taking into account variability between entities of the classes. Approaches for boxplots forecasting are constructed from multivariate statistical modeling, demonstrating the usefulness of the proposed approach through experiments with synthetic and real symbolic time series.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Computer Science, Information Systems
Nguyen Ho, Huy Vo, Mai Vu, Torben Bach Pedersen
Summary: Recent advancements in computing and data technologies have made analyzing Big Data across multiple datasets essential for extracting valuable insights. Techniques need to be fast, scalable, and assist users in prioritizing correlations across different temporal scales to focus on important relationships. AMIC is a method that uses mutual information to identify correlations in large time series, with a focus on scalability and effectiveness.
IEEE TRANSACTIONS ON BIG DATA
(2021)
Article
Computer Science, Artificial Intelligence
Lasitha S. Vidyaratne, Mahbubul Alam, Alexander M. Glandon, Anna Shabalina, Chris Tennant, Khan M. Iftekharuddin
Summary: The study introduces a novel deep cellular recurrent neural network (DCRNN) architecture to efficiently process complex multidimensional time-series data with spatial information. The cellular recurrent architecture enables location-aware synchronous processing of time-series data while ensuring efficiency in dealing with high-dimensional inputs.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Multidisciplinary Sciences
Zhiwen Tian, Ming Zhuo, Leyuan Liu, Junyi Chen, Shijie Zhou
Summary: Real-world industrial systems generate a large amount of time series data from interconnected sensors. Anomaly detection on these data is crucial for fault detection and system security. However, the lack of labeled data and the failure of current techniques in utilizing spatial and temporal dependencies pose challenges. To address this, we propose STADN, a novel Anomaly Detection Network Using Spatial and Temporal Information. STADN leverages graph attention network and long short-term memory network to capture spatial and temporal dependencies in multivariate time series data, and achieves state-of-the-art performance in detecting and locating anomalies.
SCIENTIFIC REPORTS
(2023)
Article
Computer Science, Information Systems
Alimasi Mongo Providence, Chaoyu Yang, Tshinkobo Bukasa Orphe, Anesu Mabaire, George K. Agordzo
Summary: This article discusses the hybrid dynamical schemes in multi-variable time series data inference, emphasizing the importance of various influences on the future growth of MTS, proposing the use of normalization modules to improve forecasting accuracy, and studying the application of various deep learning techniques and traditional neural networks in this field.
Article
Biochemical Research Methods
Kerian Thuillier, Caroline Baroukh, Alexander Bockmayr, Ludovic Cottret, Loic Pauleve, Anne Siegel
Summary: This study presents a novel approach to infer Boolean rules for metabolic regulation from time-series data and a prior knowledge network (PKN). By combining answer set programming and linear programming, candidate Boolean regulations that can reproduce the given data are generated. The quality of predictions depends on the available time-series data, such as kinetic, fluxomics or transcriptomics data.
Article
Computer Science, Information Systems
Erik Bollen, Rik Hendrix, Bart Kuijpers, Valeria Soliani, Alejandro Vaisman
Summary: This paper presents how transportation networks can be represented and queried using temporal graph databases and temporal graph query languages. It also showcases a real-world case of analyzing water quality over time using this approach.
ISPRS INTERNATIONAL JOURNAL OF GEO-INFORMATION
(2023)
Article
Computer Science, Theory & Methods
Fang Lv, Wei Wang, Linxuan Han, Di Wang, Yulong Pei, Junheng Huang, Bailing Wang, Mykola Pechenizkiy
Summary: This study proposes a quantitative framework for mining trading patterns of pyramid schemes from financial time series data. The framework includes the LoRSD algorithm for sequence de-noising and the Contrast TPM algorithm for mining patterns. The effectiveness of the framework is demonstrated through extensive experiments on financial data.
FUTURE GENERATION COMPUTER SYSTEMS-THE INTERNATIONAL JOURNAL OF ESCIENCE
(2022)
Article
Computer Science, Artificial Intelligence
Hyeonmo Kim, Heonho Kim, Sinyoung Kim, Hanju Kim, Myungha Cho, Bay Vo, Jerry Chun-Wei Lin, Unil Yun
Summary: Periodic pattern mining is a topic focused on mining periodic event patterns with sufficient confidence. The resulting patterns are used to predict future events and have found applications in various fields such as predicting oil price fluctuations, traffic congestion, human behavior analysis, and sensor-based data analysis. However, current data structures have limitations in terms of computing performance.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Physics, Multidisciplinary
Denis Ullmann, Olga Taran, Slava Voloshynovskiy
Summary: Time series (TS) and multiple time series (MTS) predictions are important in deep learning models, with potential applications in finance, e-commerce, NLP, medicine, and physics. The information bottleneck (IB) framework has not been explored in the context of TS or MTS analysis, and we propose a new approach using partial convolution to encode time sequences into a two-dimensional representation resembling images. Our model shows good performance in electricity production, road traffic, and solar activity data.
Article
Physics, Multidisciplinary
Charles K. Assaad, Emilie Devijver, Eric Gaussier
Summary: This study addresses the problem of learning a summary causal graph on time series with potentially different sampling rates. A new causal temporal mutual information measure is proposed, which relates to an entropy reduction principle. PC-like and FCI-like algorithms are used to construct the summary causal graph, and their efficacy and efficiency are demonstrated through evaluations on multiple datasets.