Abstract by Eric Todd
Physics and Astronomy
Modeling Crowd Noise with Machine Learning
Understanding crowd behavior is an important, yet complex problem that stretches across multiple fields of study. In this project, we focus on the problem of modeling crowd tendencies from different acoustical features of crowd noise. We use both audio and video recordings of crowd sentiment to train machine learning models. I examine some of the challenges we have encountered in approaching such a complex problem, such as isolating and extracting key acoustical events, as well as selecting machine learning techniques. I discuss my role in data collection, and in developing methods to extract key acoustical features relevant to our model, in addition to researching applicable machine learning models. Our long-term objective is to better understand acoustical signals that could predict changes in a crowd’s emotion which could in turn indicate changes and shifts in crowd behavior.