MIT EECS — Lincoln Labs Undergraduate Research and Innovation Scholar
Posterior consistency in the number of components for Dynamic World Modeling
Leslie P. Kaelbling
In many problems such as object tracking, objects can be added, removed, and moved over time. One way we can model the dynamic world is to use dependent Dirichlet process mixture models (DDPMM.) However, we believe that the models are lack of consistency in the number of clusters to the true number. In other words, the number of objects inferred from the posterior model, as more data observed, does not converge to the true value. Therefore, the number of objects inferred is inaccurate. Our goal is to point out the fallacy in the DDPMM and be able to come up with a novel model that incorporates this property.
I am mainly interested in Machine Learning and Robotics. This superUROP project under Leslie will be a great opportunity for me to explore the intersection between these two areas.