Saida Elmi, PhD

As an adjunct assistant professor at Yale University, School of Medicine, Dr. Elmi's research is focused on creating a system called The Automated Test of Embodied Cognition (ATEC), which is unique in several aspects. It is the first systematic measure of the construct of embodied cognition; it is the first to use automated administration; and it is the first to use motion capture technology and machine learning algorithms to score assessments and explore individual differences.

Dr. Elmi has also conducted research at National University of Singapore and created an intelligent transportation system to predict speed, travel time and taxi fare using data collected from taxi trajectories. These arose as a result of a collaboration with the NUS-Grab Lab. Here, there are a number of inter-related problems – predicting the speed in which vehicles are moving in a traffic network, predicting the travel time in road networks, and estimating taxi fare to travel from point X to point Y. As these problems deal with both spatial data (road networks) and temporal data (trajectory data), Dr. Elmi designed different deep learning architectures that fit the problems. In particular, she was able to adapt and integrate deep learning models (e.g., RNN, CNN, ResNet) in novel ways that resulted in superior performance (in terms of accuracy/effectiveness and efficiency) compared to existing works. Her works on speed prediction and taxi fare estimation were reported in two papers in MOBIQUIOUS’2020. In her work on travel time prediction, she proposed a transfer learning framework that exploits historical data of some source areas which are data-abundant regions to predict travel time in missing data areas (SmartCity’2020). Dr. Elmi has also developed a novel framework that identifies vehicle/driving environment-dependent factors to predict energy consumption over a road network based on historical consumption data for different vehicle types (WWW’2021).

Another Data Science-oriented area that Dr. Elmi has focused on is the design of Next Points-of-Interest (POIs) Prediction recommendation engine. Here, Saida has proposed a temporal convolutional neural network (CNN)-based architecture model called Deep-POIs that learns next POI as images. Spatio-temporal check-in dynamics are converted to images describing the time and space relations of users’ mobility behaviors. She has further proposed a novel group-recommendation approach that accommodates both individual preferences and group decisions in a joint model for the next POIs prediction. More specifically, based on influencing users in a group, she has devised a hybrid deep architecture model built with graph convolution networks and attention mechanism to extract connections between group and personal preferences and then capture the impact of each user on the group decision-making. These works were published in ICTAI 2021 and ECIR 2022.