Time Series Workshop
Time Series Workshop @ ICML 2021
July 24, 2021. Virtual Workshop

Introduction

Time series is one of the fastest growing and richest types of data. In a variety of domains including dynamical systems, healthcare, climate science and economics, there have been increasing amounts of complex dynamic data due to a shift away from parsimonious, infrequent measurements to nearly continuous real-time monitoring and recording. This burgeoning amount of new data calls for novel theoretical and algorithmic tools and insights.

The goals of our workshop are to: (1) highlight the fundamental challenges that underpin learning from time series data (e.g. covariate shift, causal inference, uncertainty quantification), (2) discuss recent developments in theory and algorithms for tackling these problems, and (3) explore new frontiers in time series analysis and their connections with emerging fields such as causal discovery and machine learning for science. In light of the recent COVID-19 outbreak, we also plan to have a special emphasis on non-stationary dynamics, causal inference, and their applications to public health at our workshop.

Time series modelling has a long tradition of inviting novel approaches from many disciplines including statistics, dynamical systems, and the physical sciences. This has led to broad impact and a diverse range of applications, making it an ideal topic for the rapid dissemination of new ideas that take place at ICML. We hope that the diversity and expertise of our speakers and attendees will help uncover new approaches and break new ground for these challenging and important settings. Our previous workshops have received great popularity at ICML, and we envision our workshop will continue to appeal to the ICML audience and stimulate many interdisciplinary discussions.

Congratulations to the workshop award winners:

Best Paper Award:
Charlotte Bunne, Laetitia Meng-Papaxanthos, Andreas Krause and Marco Cuturi. JKOnet: Proximal Optimal Transport Modeling of Population Dynamics
Honorable Mention:
Shixiang Zhu, Alexander Bukharin, Liyan Xie, Shihao Yang, Pinar Keskinocak and Yao Xie. Early Detection of COVID-19 Hotspots Using Spatio-Temporal Data
Brian Lim, Xiaoyong Jin, Rachel Redberg and Yu-Xiang Wang. Ecological Inference using Constrained Kalman filters for the COVID-19 Pandemic
Best Poster Award:
Gabriel Hope, Michael Hughes, Finale Doshi-Velez and Erik Sudderth. Prediction-Constrained Hidden Markov Models for Semi-Supervised Classification

Schedule

8:45 am - 6:05 pm (pacific time)
July 24, 2021

Morning Session


8:45 - 9:00 Opening Remarks
9:00 - 9:45 Invited Talk: Time-series in healthcare: challenges and solutions (Mihaela Van der Schaar)
9:45 - 10:30 Invited Talk: Multiscale Bayesian Modelling: Ideas and Examples from Consumer Sales (Mike West)
10:30 - 10:45 Morning Coffee Break
10:45 - 11:00 Contributed Talk: JKOnet: Proximal Optimal Transport Modeling of Population Dynamics (Charlotte Bunne)
11:00 - 11:45 Invited Talk: Quantifying causal influence in time series and beyond (Dominik Janzing)
11:45 - 12:45 Poster Session Gather.Town
12:45 - 14:30 Lunch

Afternoon Session


14:30 - 14:45 Contributed Talk: PSA-GAN: Progressive Self Attention GANs for Synthetic Time Series (Paul Jeha)
14:45 - 15:30 Invited Talk: Towards modeling raw messy time series data with latent stochastic differential equations (David Duvenaud)
15:30 - 15:45 Afternoon Coffee Break
15:45 - 10:00 Contributed Talk: Early Detection of COVID-19 Hotspots Using Spatio-Temporal Data (Shixiang Zhu)
16:00 - 16:45 Invited Talk: Online Learning with Optimism and Delay (Lester Mackey)
16:45 - 17:00 Contributed Talk: Ecological Inference using Constrained Kalman filters for the COVID-19 Pandemic (Brian Lim)
17:00 - 18:00 Poster Session Gather.Town
18:00 - 18:05 Awards and Closing Remarks
 

Keynote Speakers

David Duvenaud

Assistant Professor
University of Toronto

Towards modeling raw messy time series data with latent stochastic differential equations

Dominik Janzing

Principal Research Scientist
Amazon

Quantifying causal influence in time series and beyond

Lester Mackey

Researcher
Microsoft Research

Online Learning with Optimism and Delay

Mihaela Van der Schaar

Professor
University of Cambridge & UCLA

Time-series in healthcare: challenges and solutions

Mike West

Distinguished Professor
Duke University

Multiscale Bayesian Modelling: Ideas and Examples from Consumer Sales

Abstract
Bayesian multiscale models exploit variants of the "decouple/recouple" concept to enable advances in forecasting and monitoring of increasingly large-scale time series. Recent and current applications include financial and commercial forecasting, as well as dynamic network studies. I overview some recent developments via examples from applications in large-scale consumer demand and sales forecasting with intersecting marketing related goals. Two coupled applied settings involve (a) models for forecasting daily sales of each of many items in every supermarket of a large national chain, and (b) models for understanding and forecasting customer/household-specific purchasing behavior to informs decisions about personalized pricing and promotions on a continuing basis. The multiscale concept is applied in each setting to define new classes of hierarchical Bayesian state-space models customized to the application. In each area, micro-level, individual time series are represented via customized model forms that also involve aggregate-level factors, the latter being modelled and forecast separately. The implied conditional decoupling of many time series enables computational scalability, while the effects of shared multiscale factors define recoupling to appropriately reflect cross-series dependencies. The ideas are of course relevant to other applied settings involving large-scale, hierarchically structured time series.

Accepted Papers

Rohit Girdhar, Laura Gustafson, Aaron Adcock and Laurens van der Maaten. Forward Prediction for Physical Reasoning

Shixiang Zhu, Alexander Bukharin, Liyan Xie, Shihao Yang, Pinar Keskinocak and Yao Xie. Early Detection of COVID-19 Hotspots Using Spatio-Temporal Data

Shima Imani, Alireza Abdoli and Eamonn Keogh. Time2Cluster: Clustering Time Series Using Neighbor Information

Adèle Gouttes, Kashif Rasul, Mateusz Koren, Johannes Stephan and Tofigh Naghibi. Probabilistic Time Series Forecasting with Implicit Quantile Networks

Nathan Sesti, Juan Jose Garau-Luis, Edward Crawley and Bruce Cameron. Integrating LSTMs and GNNs for COVID-19 Forecasting

Juliane Weilbach, Sebastian Gerwinn, Christian Weilbach and Melih Kandemir. Inferring the Structure of Ordinary Differential Equations

Paul Jeha, Michael Bohlke-Schneider, Pedro Mercado, Rajbir Singh Nirwan, Shubham Kapoor, Valentin Flunkert, Jan Gasthaus and Tim Januschowski. PSA-GAN: Progressive Self Attention GANs for Synthetic Time Series

Joel Dyer, Patrick Cannon and Sebastian Schmon. Deep Signature Statistics for Likelihood-free Time-series Models

Ludovico Giorgini, Soon Hoe Lim, Woosok Moon, Nan Chen and John Wettlaufer. Modeling the El Niño Southern Oscillation with Neural Differential Equations

Duc Nguyen, Binh Nguyen, Phuoc Nguyen and Truyen Tran. High-Order Representation Learning for Multivariate Time Series Forecasting

David Blanco-Mulero, Markus Heinonen and Ville Kyrki. Evolving-Graph Gaussian Processes

Rishab Guha, Eric Aldrich, Bertram Ieong, Shanshan Li, Domenico Giannone and Juan Huerta. Towards Robust, Scalable and Interpretable Time Series Forecasting using Bayesian Vector Auto-Regression

Linsey Pang and Ketki Gupte. Robust Price Optimization in Retail

Genevieve Flaspohler, Francesco Orabona, Judah Cohen, Soukayna Mouatadid, Miruna Oprescu, Paulo Orenstein and Lester Mackey. Online Learning with Optimism and Delay

Clayton Rooke, Jonathan Smith, Kin Kwan Leung, Maksims Volkovs and Saba Zuberi. Temporal Dependencies in Feature Importance for Time Series Predictions

Brian Lim, Xiaoyong Jin, Rachel Redberg and Yu-Xiang Wang. Ecological Inference using Constrained Kalman filters for the COVID-19 Pandemic

Zhen Wang, Yang Zhang, Ai Jiang, Ji Zhang, Zhao Li, Jun Gao, Ke Li and Chenhao Lu. DAMA-Net: A Novel Predictive Model for Irregularly Asynchronously andSparsely Sampled Multivariate Time Series

Ji Won Park, Ashley Villar, Yin Li, Yan-Fei Jiang, Shirley Ho, Joshua Yao-Yu Lin, Philip Marshall and Aaron Roodman. Inferring Black Hole Properties from Astronomical Multivariate Time Series with Bayesian Attentive Neural Processes

Daniel Zügner, Francois-Xavier Aubet, Victor Garcia Satorras, Tim Januschowski, Stephan Günnemann and Jan Gasthaus. A Study of Joint Graph Inference and

Francois-Xavier Aubet, Daniel Zügner and Jan Gasthaus. Monte Carlo EM for Deep Time Series Anomaly Detection

Charlotte Bunne, Laetitia Meng-Papaxanthos, Andreas Krause and Marco Cuturi. JKOnet: Proximal Optimal Transport Modeling of Population Dynamics

Gabriel Hope, Michael Hughes, Finale Doshi-Velez and Erik Sudderth. Prediction-Constrained Hidden Markov Models for Semi-Supervised Classification

Dheeraj Baby, Hilaf Hasson and Bernie Wang. Revisiting Dynamic Regret of Strongly Adaptive Methods

Cristian Challu, Kin G. Olivares, Gus Welter and Artur Dubrawski. DMIDAS: Deep Mixed Data Sampling Regression for Long Multi-Horizon Time Series Forecasting

Ruizhi Deng, Marcus Brubaker, Greg Mori and Andreas Lehrmann. Continuous Latent Process Flows

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 2021 style format. Supplementary material is permitted, although there is no guarantee that it will be reviewed. The review process is double blind, and authors should follow the same submission guidelines as for the main conference (see https://icml.cc/Conferences/2021/CallForPapers). All inquiries may be sent to tswicml2021@gmail.com.

All accepted papers will be presented in a virtual poster session, and some will be selected for oral presentation. We will also select best paper awards based on scientific merit, impact, and clarity. A $300.00 USD cash prize will be awarded to the 1st prize best paper. Best paper awards are nominated by program committee and judged by the Best Paper award committee.

Submissions page: ICML 2021 Times Series Workshop.

Note on dual submission and publication policy: We accept dual submissions with other conferences, workshops, and/or journals. We welcome articles currently under review or papers planned for publication elsewhere. However, papers that have been published at an ML conference should not be submitted. Accepted papers will be published on the TSW homepage, but are to be considered non-archival.

Paper Submission Deadline: June 4 (extended from June 1), 2021, 11:59 PM PST

Author Notification: June 27 (previously June 17th), 2021, 11:59 PM PST

Final Version: July 9th (previously June 27th), 2021, 11:59 PM PST

Workshop: July 24, 2021

Workshop Organizers

Yian Ma

UC San Diego

Ehi Nosakhare

Microsoft

Yuyang Wang

Amazon Web Services

Scott Yang

D. E. Shaw & Co.

Rose Yu

UC San Diego

Sponsored by