International Workshop on Temporal Analytics

The rapid advancement in sensing technologies and computational power have led to an unprecedented growth of data, especially temporal data, or what is also commonly known as time series data. They are ubiquitous and are now seen in almost all applications including remote sensing, medicine, finance, engineering, smart cities and many more. Temporal analytics is an important field that has been studied extensively for the last decade, with hundreds to thousands of papers being published each year in various domains. Despite the advancement of new approaches and superiority of state-of-the-art algorithms, modern time series data continue to pose significant challenges to existing approaches. Some of these challenges include the high dimensionality of the data (in terms of number of variables and timesteps), noisiness and irregularity of the data. Additionally, most existing algorithms are not explainable, do not scale and require high-quantity of labelled data, which we do not always have access to. The importance of scalability becomes more prominent as the number of data increases. Therefore, designing new approaches and solutions that are able to tackle these challenges is critical.

The objective of this workshop is to bring researchers in this area to discuss new and existing challenges in temporal analytics, which covers a wide range of tasks including classification, regression, clustering, anomaly detection, retrieval, feature extraction and learning representations. The solutions can be algorithmic, theoretical or systems-based in nature.

This workshop will discuss a wide variety of topics including but are not limited to:

Time series analysis tasks such as classification, extrinsic regression, forecasting, clustering, annotation, segmentation, anomaly detection and pattern mining
Time series analysis with no or few supervision
Early classification of time series
Deep learning for time series
Learning representation for time series
Modelling temporal dependencies
Spatial-temporal statistical analysis
Functional data analysis methods
Scalable time series analytics methods
Explainable time series analysis methods
Time series that are sparse, involve irregular sampling
Time series with missing values and variable lengths
Multivariate time series with high dimensionality and heterogenous
Interdisciplinary time series analysis applications

Important Dates

Paper submission deadline: ~~March 7, 2023~~March 21, 2023

Paper acceptance notification: ~~April 7, 2023~~April 21, 2023

Paper camera-ready deadline: ~~April 25, 2023~~May 2, 2023

Workshop day: May 25, 2023

*All deadlines are 23:59 Pacific Standard Time (PST)

Submissions

Paper submission must be in English. All papers will be double-blind reviewed by the Program Committee based on technical quality, relevance to data mining, originality, significance, and clarity. All paper submissions will be handled electronically. Papers that do not comply with the Submission Policy will be rejected without review.

Each submitted paper must include an abstract up to 200 words and be between 6-12 single-spaced pages with 10pt font size (including references, appendices, etc.). Authors must use Springer LNCS/LNAI manuscript submission guidelines for their submissions. All papers must be submitted electronically through the CMT paper submission system in PDF format only. Supplementary material may be submitted as a separate PDF file, but reviewers are not obligated to consider this, and your manuscript should, therefore, stand on its own merits without any supplementary material. Supplementary material will not be published in the proceedings. Each accepted paper will be invited to propose a camera ready version of their article taking into account the reviewers recommendations.

The submitted papers must not be previously published anywhere but may be under consideration by any other conference or journal during the workshop review process. Submitting a paper to the conference means that if the paper was accepted, at least one author will complete the regular registration and attend the workshop to present the paper.

Submission website:

https://cmt3.research.microsoft.com/IWTA2023/

Program

This workshop will be held in conjunction with the 27th Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 2023) on the 25th May 2023. It will be organized fully online.

0900-0910:

Workshop introduction

0910-0930:

Zheng Sun, Yi Wei, Wenxiao Jia and Long Yu
Stecformer: Spatio-temporal Encoding Cascaded Transformer for Multivariate Long-term Time Series Forecasting

0930-0950:

Abdellah Madane, Mohamed Djallel DILMI, Florent Forest, Hanane Azzag, Mustapha Lebbah and LACAILLE Jerome
Transformer-based Conditional Generative Adversarial Network for Multivariate Time Series Generation

0950-1010:

Trent Henderson, Annie G Bryant and Ben D Fulcher
Never a Dull Moment: Distributional Properties as a Baseline for Time-Series Classification

1015-1115:

Invited talk by Professor James Bailey, The University of Melbourne
Backdoor Attacks on Time Series: A Generative Approach

1115-1130:

Workshop closing

*All times are given in JST (UTC+9, Osaka time).

Keynote Speakers

Professor James Bailey, The University of Melbourne

james_bailey

Title: Backdoor Attacks on Time Series: A Generative Approach

Abstract:

Backdoor attacks have emerged as one of the major security threats to deep learning models as they can easily control the model’s test-time predictions by pre-injecting a backdoor trigger into the model at training time. While backdoor attacks have been extensively studied on images, few works have investigated the threat of backdoor attacks on time series data. To fill this gap, we propose a novel generative approach for time series backdoor attacks against deep learning based time series classifiers. Backdoor attacks have two main goals: high stealthiness and high attack success rate. We find that, compared to images, it can be more challenging to achieve these two goals on time series. This is because time series have fewer input dimensions and lower degrees of freedom, making it hard to achieve a high attack success rate without compromising stealthiness. Our generative approach addresses this challenge by generating trigger patterns that are as realistic as realtime series patterns while achieving a high attack success rate without causing a significant drop in clean accuracy. We also show that our proposed attack is resistant to potential backdoor defenses. Furthermore, we propose a novel universal generator that can poison any type of time series with a single generator that allows universal attacks without the need to fine-tune the generative model for new time series datasets.