Time Series Workshop
Time Series Workshop @ ICML 2019
June 14, 2019. Long Beach, USA

Introduction

Time-dependent sequential data emerge in many key real-world problems, including areas such as climate, robotics, biology, economics, entertainment, healthcare and transportation. The increasing volume and complexity of time series data in modern applications highlight the importance of scalable and flexible time series learning techniques. Predominant methods in machine learning often assume i.i.d. observations, which is generally not appropriate for time series data. Therefore, there is both a great need and an exciting opportunity for the machine learning community to develop theory, models, and algorithms for processing and analyzing large-scale complex time series data.

The Time Series Workshop at ICML 2019 brings together theoretical and applied researchers at the forefront of time series analysis and algorithms to discuss existing key progress and promising new directions. Topics include methods for time series prediction, classification, clustering, anomaly and change point detection, causal discovery, and dimensionality reduction as well as a general theory for learning and analyzing stochastic processes. Our workshop will host leading researchers from academia and industry with a range of perspectives and interests. We also invite researchers from the related areas of batch and online learning, deep learning, reinforcement learning, data analysis and statistics, and many others to both contribute and participate in this workshop.

Due to our generous sponsors, we have a limited amount of funding to support the travel of graduate students and postdoctoral fellows (up to two years) to attend our workshop. These awards will be assigned primarily based on need and travel distance, and they will be limited to one individual per paper. If you or one of your authors would like to be considered for a travel award, please submit the following information to tsw.icml2019@gmail.com no later than May 28th, 2019 11:59PM Pacific Time. 1) Workshop paper title; 2) Student / postdoctoral fellow name; 3) Student / postdoctoral fellow level; 4) Traveling from; 5) Reasons for needing support. The recipients of these travel awards will be notified by June 6th (updated from June 4th), 2019 11:59 PM Pacific Time.

Schedule

Long Beach Convention Center, Long Beach
8:45 am - 6:05 pm
June 14, 2019

Morning Session


8:45 - 9:00 Opening Remarks
9:00 - 9:45 Invited Talk: Yulia Gel, Change Point Detection in Time Series through the Lens of Topological Data Analysis
9:45 - 10:00 Contributed Talk: Online Forecasting of Total-Variation-bounded Sequences (Dheeraj Baby and Yu-Xiang Wang)
10:00 - 10:15 Contributed Talk: Latent Ordinary Differential Equations for Irregularly-Sampled Time Series (Yulia Rubanova, Ricky T. Q. Chen and David Duvenaud)
10:15 - 10:30 Contributed Talk: BreizhCrops: A Satellite Time Series Dataset for Crop Type Identification (Marc Rußwurm, Sébastien Lefèvre and Marco Körner)
10:30 - 11:00 Morning Coffee Break
11:00 - 11:45 Invited Talk: Suchi Saria, Addressing Failures from Feedback Loops in Designing Decision Aids
11:45 - 12:30 Poster Session
12:30 - 14:30 Lunch

Afternoon Session


14:30 - 14:45 Contributed Talk: Neural time series models with GluonTS (Alexander Alexandrov et. al.)
14:45 - 15:00 Contributed Talk: Latent Spectrum Gaussian Processes (Jayson Salkey, Greg Benton, Wesley Maddox, Julio Albinati and Andrew Wilson)
15:00 - 15:30 Panel Discussion
15:30 - 16:00 Afternoon Coffee Break
16:00 - 16:30 Poster Session
16:30 - 17:15 Invited Talk: Yan Liu, Artificial Intelligence for Smart Transportation
17:15 - 18:00 Invited Talk: Edo Liberty, Streaming algorithms, Apache DataScketches, and new results on corsets
18:00 - 18:05 Awards and Closing Remarks
 

Keynote Speakers

Yulia Gel

Professor
University of Texas at Dallas

Change Point Detection in Time Series through the Lens of Topological Data Analysis

Abstract
While geometrical methods continue to gain popularity in statistical sciences and machine learning, from causal inference to deep learning on manifolds, the role of geometry in change point analysis of time series and temporally dependent processes remains yet largely unexplored. In this project we advocate that information on the inherit data shape can provide an invaluable insight into the hidden changes in the data structure, organization, and dynamics of temporally dependent data. In particular, our approaches are inspired by the principle “Data has Shape, Shape has Meaning” -- the driving motto of Ayasdi software for topological data analysis, which we re-phrase as “Data has Shape, Look for Changes in Shape, They might Signal Threats”. To study latent data shape, we invoke machinery of persistent homology that allows us to systematically infer qualitative and quantitative geometric and topological structures from the time series data. The key rationale of our methodology is then to identify changes in the data generating process through locating the associated changes in the time series of geometric and topological structures characterized by persistent homology. We illustrate the proposed methodology in application to environmental, financial and power system data.

Suchi Saria

Assistant Professor
Johns Hopkins University

Addressing Failures from Feedback Loops in Designing Decision Aids

Abstract
Coming Soon

Yan Liu

Associate Professor
University of Southern California

Artificial Intelligence for Smart Transportation

Abstract
Data-enabled smart transportation has attracted a surge of interest from machine learning and artificial intelligence researchers nowadays due to the bloom of sensing technologies and online ride-hailing industry. Large-scale high quality route data and trading data (spatiotemporal data) have been generated every day, which makes AI an urgent need and preferred solution for the decision making in intelligent transportation systems. While a large of amount of work have been dedicated to traditional transportation problems, they are far from satisfactory for the rising need. In this talk, I will discuss the key challenges in the areas, and our recent work on devising AI solutions to these challenges.

Edo Liberty

Founder
HyperCube

Streaming algorithms, Apache DataScketches, and new results on corsets

Abstract
Modern data platforms often observe data at such a rate that storing it for future analytics is unfeasible. Therefore, techniques were developed to analyze streams of data on the fly using only a fixed memory footprint (without storing the data). Such techniques are called streaming algorithms or sketching. This talk will explain the streaming algorithms' data model, its benefits, and challenges. The talk will introduce the Apache DataSketches library and some of its functionality. Finally, we will cover new results about coresets. These results indicate that small sketches exist for problems like density estimation, regression, classification, and many other tasks.

Accepted Papers

Edvin Listo Zec, Henrik Arnelid and Nasser Mohammadiha. Recurrent Conditional GANs for Time Series Sensor Modelling

Ruofeng Wen and Kari Torkkola. Deep Generative Quantile-Copula Models for Probabilistic Forecasting and Simulation

Saurabh Agrawal, Saurabh Verma, Anuj Karpatne, Stefan Liess, Snigdhansu Chatterjee and Vipin Kumar. A Fast-Optimal Guaranteed Algorithm For Learning Sub-Interval Relationships in Time Series

Xiaoyong Jin, Shiyang Li, Yunkai Zhang and Xifeng Yan. Multi-step Deep Autoregressive Forecasting with Latent States

Shenda Hong, Cao Xiao, Trong Nghia Hoang, Tengfei Ma, Hongyan Li and Jimeng Sun. RDPD: Rich Data Helps Poor Data via Imitation

Ivan Kiskin, Udeepa Meepegama and Stephen Roberts. Super-resolution of Time-series Labels for Bootstrapped Event Detection

Tianhao Zhu and Sergul Aydore. Time-Smoothed Gradients for Online Forecasting

Alexander Alexandrov, Konstantinos Benidis, Michael Bohlke-Schneider, Valentin Flunkert, Jan Gasthaus, Tim Januschowski, Danielle Maddix, Syama Rangapuram, David Salinas, Jasper Schulz, Lorenzo Stella, Ali Caner Türkmen and Yuyang Wang. Neural time series models with GluonTS

Marc Rußwurm, Sébastien Lefèvre and Marco Körner. BreizhCrops: A Satellite Time Series Dataset for Crop Type Identification

Kezi Yu, Yunlong Wang, Yong Cai, Cao Xiao, Emily Zhao, Lucas Glass and Jimeng Sun. Rare Disease Detection by Sequence Modeling with Generative Adversarial Networks

Fan Zhang, Tong Wu, Yunlong Wang, Yong Cai, Cao Xiao, Emily Zhao, Glass Lucas and Jimeng Sun. Predicting Treatment Initiation from Clinical Time Series Data via Graph-Augmented Time-Sensitive Model

Tailin Wu, Thomas Breuel, Michael Skuhersky and Jan Kautz. Nonlinear Causal Discovery with Minimum Predictive Information Regularization

Saelig Khattar, Hannah O'Day, Paroma Varma, Jason Fries, Jen Hicks, Scott Delp, Helen Bronte-Stewart and Chris Re. Multi-frame Weak Supervision to Label Wearable Sensor Data

Yingxiang Yang, Niao He and Negar Kiyavash. Learning Poisson Intensities with Pseudo Mirror Descent

Anish Mathew, Deepak P and Sahely Bhadra. Warping Resilient Time Series Embeddings

Jayson Salkey, Greg Benton, Wesley Maddox, Julio Albinati and Andrew Wilson. Function-space Distributions over Kernels

Kun Zhang, Yuan Xue, Gerardo Flores, Alvin Rajkomar, Claire Cui and Andrew Dai. Time series modelling by restricting feature interaction

Call for Papers

We invite researchers to submit both theoretical and applied work on time series analysis, modeling, and algorithms, along with their applications. Papers submitted to the workshop should be up to four pages long excluding references and in ICML 2019 format. Supplementary material is permitted, although there is no guarantee that it will be reviewed. As the review process is not blind, authors can reveal their identity in their submissions. All inquiries may be sent to tswicml2019@gmail.com.

Submissions page: Times Series Workshop 2019.

Note on open dataset submissions: In order to promote new and innovative research in time series research, we plan to accept a small number of high quality time series dataset contributions. These submissions should be accompanied by a clear and detailed description of the dataset, some potential questions and applications that arise from it, as well as discussion on why the data cannot be sufficiently modeled using traditional batch learning techniques. Preliminary empirical investigations conveying any insight about the data will increase the quality of the submission.

Note on dual submission and publication policy: We accept dual submissions with other conferences, workshops, and/or journals. Accepted papers will be posted on the workshop website. However, there will not be any formal proceedings, so authors should feel free to submit their accepted work to other venues in the future (subject to the submission policy of those venues).

Key Dates

 

Paper Submission Deadline: May 3, 2019, 11:59 PM PST

Author Notification: May 20, 2019, 11:59 PM PST (changed from May 17, 2019, 11:59PM PST)

Travel Award Application Due: May 28, 2019, 11:59 PM PST

Travel Award Notification: June 6 (changed from June 4th), 2019, 11:59 PM PST

Final Version: June 13, 2019, 11:59 PM PST

Workshop: June 14, 2019

Workshop Organizers

Vitaly Kuznetsov

Google Research

Cheng Tang

Amazon Web Services

Yuyang Wang

Amazon Web Services

Scott Yang

D. E. Shaw & Co.

Rose Yu

Northeastern University

 

Sponsored by