Machine learning in data assimilation

Abstract: Data assimilation is concerned with sequentially estimating a time-evolving signal from partial and noisy observations. The signal is often assumed to evolve according to a known dynamical system, but in practice, the dynamics are rarely fully known and are subject to significant model error. Machine learning techniques have recently emerged as a promising avenue to improve dynamical system models for data assimilation in important applications such as numerical weather forecasting. In addition to alleviating model error, machine-learned models are cheaper to evaluate and hence enable running multiple parallel forecasts for uncertainty quantification. Machine learning techniques also provide several other potential advantages, including the online co-learning of the signal and the dynamics, the improvement of the spatial resolution of signal estimates, and the data-driven enhancement of data assimilation algorithms. This talk will survey some recent advances and open questions in combining machine learning with data assimilation.

Speaker Name and Bio: Daniel Sanz-Alonso, Associate Professor at University of Chicago

Host: Olivia Martin, Justin Shea, Mehdi Jeddi, and Sou-Cheng Choi

Talk Format: This is a hybrid event. To attend online, join us on Zoom here at 6pm:
https://numfocus-org.zoom.us/j/89399976851?pwd=UEgMUZXdYmKdK1x1dIPL6hwUYnp7NW.1

Sponsor: Adyen, UIC College of Business, and PyData Chicago co-host this event. UIC will provide the meeting site. Adyen will sponsor pizza and soft drinks for the onsite participants.

Address: University of Illinois - Chicago, Douglass Hall, Room 220, 705 S Morgan St, Chicago, IL 60607

Logistics: “UIC Douglass Hall” is recognized on Google Maps, which can guide you through campus. Once you arrive, proceed to the second floor, room number 220