Time series is one of the fastest growing and richest types of data. In domains spanning as broad a range as epidemiology, climate science, healthcare, and transportation, there has been an increasing amount of complex temporal data due to a shift away from parsimonious, infrequent measurements to nearly continuous real-time monitoring and recording. This has caused a drastic increase in the size and complexity of time series data, calling for novel theoretical, algorithmic and modeling tools in machine learning.

The recent COVID-19 outbreak is one of the most impactful and challenging problems in our world data. It has also highlighted the importance and difficulties of time series forecasting in a setting with non-stationary dynamics, few sparse observations, potentially complemented by exogenous data sources. For instance, an accurate forecast of the next epicenter from just a few historic observations (complemented with social media) may allow for more efficient planning and allocation of medical resources. Challenges of similar nature go beyond epidemiological forecasting and similar problems arise in many other domains (e.g. demand forecasting of new products, new user engagements).

We plan to have a special emphasis on non-stationary dynamics, few-shot learning, and exogenous data sources for time series forecasting problems at our workshop by inviting active researchers in this field as invited speakers and making it a primary focus during our panel discussion. By bringing together researchers from different backgrounds and with different training, our workshop aims to highlight the generality and broad applicability of this problem, clarify the essential questions that underpin it (e.g. domain adaptation, non-stationarity), and to discuss various approaches for tackling it. We hope that the diversity and expertise of our speakers and attendees will help uncover new approaches and break new ground for this challenging setting.

Unlike many other domains and workshops, time series modelling has a long tradition of inviting novel approaches from many disciplines including finance, statistics, control and dynamical systems, physical sciences, leading to a broad impact with a diverse range of applications. This makes it an ideal topic for the informal discussions and rapid dissemination of new ideas that take place at a NeurIPS workshop.

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 NeurIPS 2020 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 tswicml2020@gmail.com.

Submissions page: Times Series Workshop 2020.

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).