Research

Multi-agent systems are groups of individuals whose collective behaviour allows them to complete some task. For example, a swarm of bees might work together to locate a promising food source, cars/bikes on the road follow specific rules to get to their individual desired locations, or fans in a sports stadium will work together to do the wave at a football game. Researchers use graph neural networks to perform trajectory prediction (given one time step), predict the next time step in a multi-agent system, though they typically do not verify that their model captures the appropriate individual behaviour. My work is to verify which models capture this behaviour appropriately and propose models that capture this behaviour.

Viscous fluids flow down an inclined surface with relatively predictable behavior. When we add two kinds of particles into the fluid (thus making the mixture a bidensity particle-laden slurry) is this behaviour still predictable? The answer is yes and we experimentally verified the following behaviours: (1) the front of the fluid, regardless of the particles, travels at a velocity asymptotic to t^1/2, where t is time, (2) the particles settle separately at low particle concentrations and incline angle, (3) the particles form a ridge at the front of the fluid at high particle concentration and incline angle.

Large cities, like Los Angeles, have large homeless populations and thus invest significant resources to support these populations. Modeling and understanding the dynamics of these populations allows us to properly allocate these resources. With a team of researchers at UCLA, we worked to examine if machine learning methods could help us (1) find relationships between homeless populations and other area-specific variables (like number of bus stops and average rent in an area) and (2) forecast homeless populations in future years. We found success in shallow neural networks for forecasting, and non-negative matrix factorization and correlation analysis for finding important relationships.