ISSP 2030 Spring AI Forum Week 14

21 Apr 2021 - Arun Balajiee

Date: 04/21/2021

Speaker: Neil Munjal

Title: Machine Learning for Neurological Injury in the Pediatric ICU: Promises and Challenges

In this presentation Neil talked about the application of Machine Learning (ML) in the field of medicine and as use for clinicians. The problem he and his team are targetting is to predict mortality in children which has been found to be important increasingly in the Pediatric ICUs (PICU). There have been models used in the PICUto predict mortality which could make 4-12 hour predictions on mortality, but there are no models that are good at predicting mortality or modbitdity with neurological injury in the PICU. So the problems solved by Neil and colleagues is to predict the severity of illness, assessing quality of care across sites, and ridging real-time monitoring. Specifically, the goals were to focus on population-level analysis to rperesent severity of illness and cross-site quality of are comparison. Then NEil quickly transitioned to explaining the interpretability of differentm odels such as Random Forests, Logistic Regression and others. Neural Networks were not known to perform well in at the task.

Neil then talked about causal discovery and his work in the space of causal discovery, handling missing data and other similar explorations in the space.

Key takeaways from the talk were things to consider in terms of model performance, execution speed, model explainability, non-linear modeling for solutions to problems

Speaker: Arun Balajiee

Title: Question Generation for Dialogic Reading

In this presentation, Arun talked about dialogic reading, a practice where parents and children engage in a dialogue while reading stories. He then talked about the implementation of a technological intervention that assists parents with dialogic reading. More on his work can be found in the publication.

Key takeways from the talk were things to be considered in terms of applying principles of NLP in an human computer interaction task.