Project: Explainable AI for Event and Time Series Data to Understand Neonatal Sepsis Predictions in Collaboration with the Hospital
Description
Prematurely born
babies who are admitted to the NICU (neonatal intensive care unit) in hospitals
could develop sepsis. This is threatening for their health, therefore,
researchers try to predict sepsis based on multidimensional event and time
series data collected while monitoring these babies. This project is in
collaboration with professor Carola van Pul from the applied physics department
and the MMC hospital in Veldhoven and PhD candidate Zheng Peng from the
electrical engineering department. They developed/evaluated a ResNet model with
attention layers to predict sepsis. However, we would like to open the black
box and understand the model and the predictions better using visual analytics.
By using visual analytics, users visually explore this black box and by
combining their domain knowledge with interactions reach insights about this
black box. Possible directions based on your interests could include
exploration of the parameter space and how this influences model results or
trying to understand what happens inside the model. If you are interested in
this project please contact Sanne van der Linden.
Details
- Supervisor
-
Sanne van der Linden
- Secondary supervisor
-
Stef van den Elzen
- Interested?
- Get in contact