How to Turn Confrontations in the Classroom into Teachable Moments
This project aims to develop opportunities for teachers to practice dialogue techniques in realistic but safe, virtually simulated environments. Rather than forcing young teachers to first encounter these conflicts in real situations, Adewole and Bywater will build a simulator to enable teachers to practice having difficult conversations using immersive 3D virtual reality. The system will create realistic settings that involve conversations between the teacher and a diverse group of artificially intelligent virtual students.
Therapist in Your Pocket: Reinforcement Learning Approaches for Contextual, Personalized Emotion Regulation Recommendations
In this project, we will learn and evaluate adaptive emotion regulation (ER) strategies for socially anxious individuals by developing methods that combine network analysis with reinforcement learning in an off-policy setting. This interdisciplinary collaboration between psychology and engineering permits a deeper understanding of the dynamics of ER in real life. By bringing machine learning tools to a big dataset of ER “in the wild,” this project has the potential to create more effective eHealth tools to provide personalized mental health services on a large scale at low cost, helping to alleviate the burden of mental illness.
Improving the Diagnosis of Parkinson's Disease through Graphical Modeling of Brain MRI
This project will explore the use of MRI methodology to evaluate patterns of pathological spread in Parkinson's disease (PD) and the application of those patterns as part of a new MRI-based staging schema for PD. Blair and Sekhon will utilize advanced machine learning methodology coupled with multi-modal magnetic resonance imaging of living PD patients to achieve this aim. Evidence contributing to understanding patterns of neuronal loss in PD patients and the development of new staging schema will improve diagnostic accuracy as well as clinical outcomes in PD.
Evaluating air pollution exposure and environmental justice by integrating novel high-resolution nitrogen dioxide and human activity datasets
Angelique Demetillo (Environmental Sciences) and Cho Jiang (Urban and Environmental Planning)
This project aims to advance knowledge of air pollution exposure by combining multiple novel high-resolution datasets. In particular, very-fine-scale atmospheric measurements of nitrogen dioxide (NO2) and human movement data for thousands of individuals in the polluted city of Los Angeles, California. Demetillo and Jiang hope to produce observationally-derived air quality exposure estimates at human spatial and temporal scales, quantifying air pollution exposure burdens with unprecedented precision.
Women and Cyber Recruitment of Extremist Groups: A Text Analysis Approach
This projects will explore how terrorist groups' online recruiting efforts differ between men and women. Through text analysis of recruiting and propaganda tweets, and recruiting information placed on Telegraph (an increasingly important recruitment tool for ISIS), Heidarysafa, Kowsari, and Odukoya hope to study how this and related organizations draw in women. Understanding this variance in recruiting methodology would allow intervention in a more targeted way during the recruitment process, since most existing techniques are tailored toward men. This is an important and timely correction since female recruits are increasingly used to attack soft targets like markets and schools.
The Effect of Alarm Modalities on Drivers’ Performance in Dangerous Scenarios
Nauder Namaky (Psychology) and Erfan Pakdamanian (Systems and Information Engineering)
This project will explore the development of a novel methodology to predict critical situations and optimally alert autonomous vehicle drivers such that they are able to respond in a timely and informed fashion. Namaky and Pakdamanian will develop an adaptive deep neural learning algorithm that takes into account drivers’ cognitive states in conjunction with external hazards. Using this, they will develop an alarm system that leverages active physiological data from a driver to tailor alarm type (e.g. haptic, visual, auditory or combination based on the effectiveness) in order to best shift drivers into an engaged cognitive state without causing anxiety that may impair driving performance.