SSA ACT invites everyone to attend its August Branch Meeting, where Dan Pagendam from CSIRO Data61 will be dazzling us with hybrid machine learning and statistical models for spatio-temporal prediction!
Details of the talk and its Zoom link are given below.
Date: Thursday 29 August 2024
Time: Starts 5:45pm and finishes by 7pm
Speaker: Dan Pagendam (CSIRO Data61) [Joint work with Sreekanth Janardhanan (CSIRO Environment), Joel Dabrowski (CSIRO Data61), Dan Mackinlay (CSIRO Data61)]
Venue: Room 3.02 in Marie Reay Teaching Centre, Australian National University (MRTC ANU), or via Zoom with details below.
https://anu.zoom.us/j/84738141969?pwd=LXDeHwo4mp2RkWw5TgjU7KPaPdTr7U.1
Meeting ID: 847 3814 1969
Password: 865353
Topic: A Neural-Statistical Hybrid Model for Spatio-temporal Prediction of Groundwater Dynamics in Bangladesh
Abstract: Deep neural networks are powerful models capable of learning useful representations from large complex datasets for the purpose of prediction. Such models may offer great potential in applied statistical applications if they can be assembled into architectures that facilitate interpretability and that can reliably quantify uncertainty. In this talk such an application, where observed groundwater levels at monitoring wells in the Indo-Gangetic Basin, Bangladesh, were modelled using readily available spatial and spatio-temporal predictors. We constructed a model with two deep neural sub-architectures embedded within a log-additive statistical model. The sub-architectures allowed for the partitioning of groundwater dynamics arising from: (i) local hydrological forcings; and (ii) broad-scale spatio-temporal trends. The separability of these two components facilitated interpretation of the dominant groundwater trends in the basin. The log-additive statistical model also allowed for the quantification of uncertainty in predictions of groundwater levels. We observed good coverage of out-of-sample data, indicating that uncertainty quantification from the deep neural statistical model was reliable. From an application standpoint, we demonstrate that groundwater dynamics can be reliably and easily predicted at large spatial and temporal scales that may aid governments in managing water resources. From a methodological perspective, we demonstrate that it is possible to structure models to harness the powerful attributes of deep neural networks, whilst retaining some of the desirable properties of statistical models.
Biography: Dan Pagendam is a Team Leader and Senior Research Scientist at the Commonwealth Scientific and Industrial Research Organisation (CSIRO), Australia. His research interests are in Bayesian statistical modelling, physical statistical modelling, uncertainty quantification, deep learning, and model emulation. Dan's research draws on the use of complex models and datasets to answer important scientific questions in a diverse range of fields including population biology, hydrology, and agriculture. In recent years, Dan's research has focused on the use of statistics and machine learning for biological control of invasive mosquito populations, prediction of groundwater dynamics, wildfire modelling, and quantifying uncertainty surrounding predictions of soil carbon sequestration in agriculture.
Dinner: After the talk we will be holding a dinner at 7.15pm at Taj Mahal Indian Restaurant, 39 Northbourne Ave, Canberra (https://tajmahalindianrestaurant.com.au/ )
If you are interested in attending the dinner, please let us know by 5pm Wednesday 28 August by entering your details at SSA Canberra Branch dinner attendance sheet or contacting Warren Muller (warren.muller@csiro.au ; 0407 916 868). Please regard this as a firm commitment, not just an intention. For withdrawals after the deadline, please remove your name from the sheet and phone or text Warren (0407 916 868).
NOTE: We are offering discounts to SSA early career and student members who attend dinner! For this meeting, dinners will be a fixed charge of $10 for student members and $20 for early career members.
@StatSocAus
