Mitchell Lecture: Some Examples at the Interface of “AI” and Dynamic Statistical Models for Complex Environmental and Ecological Data
Professor Christopher Wikle (University of Missouri)
Tuesday 16th February, 2021 16:00-17:00 Zoom
What do environmental processes such as those that control seasonal precipitation or temperature variation have in common with ecological processes such as the settling patterns of migratory waterfowl, or the decisions that lead to collective animal movement? They are all complex processes that included linear and nonlinear interactions across multiple variables and multiple spatial and temporal scales of variability. Such processes can be modeled statistically via dynamic spatio-temporal models (DSTMs). Yet, it is particularly challenging to specify parameterizations for complex DSTMs that are simultaneously useful scientifically, efficient computationally, and allow for proper uncertainty quantification. In some cases, when such information is available, one can embed mechanistic information into multi-level (deep/hierarchical) models to facilitate parameter reduction and interpretability. When such information is not available (such as environmental processes at very short or long time scales, or with animal behavior) then alternative learning strategies such as deep neural models, reservoir learning, or reinforcement learning can be applied. One challenge with such methods can be uncertainty quantification and the necessity of having very large data sets for model training. Here I present some recent examples where we have integrated these “AI” methods in statistical frameworks to accommodate uncertainty quantification, with the choice of methodological approach depending on data type and availability. Examples will be presented from ocean and atmospheric science and ecology that illustrate the successful blending of statistical models with transfer learning, reinforcement learning, and reservoir learning.
To attend, please register at https://www.eventbrite.co.uk/e/mitchell-lecture-2021-registration-139662096089