The objective of this project is to develop algorithms that can predict future events, in the context of a data stream containing multiple correlated threads with many types of patterns. The domain chosen to explore this research is music. Music is richly patterned, with long-term dependencies, dependencies across time-scales, and correlations between parallel information streams. This makes it an ideal domain for advancing predictive modeling in information streams. When a person is listening to a song, the listener is anticipating, at any given moment, the timing and nature of the next event by decoding the musical signal. Even when analyzing a simple song, the brain utilizes complex correlations between the musical elements to make accurate predictions.
The research will examine the following specific questions: (1) Can prediction be improved by using ensemble methods, such as mixtures-of-experts and product-of-experts, in which the predictions of multiple models are merged? (2) Can novel Probabilistic Graphical Models (PGMs) improve upon standard approaches, such as Hidden Markov Models (HMMs), for predicting events in a musical signal? (3) Can computational predictive models provide testable cognitive hypotheses that help explain how humans form mental schemas of music? A primary education goal of this proposal is to create a series of interactive online learning modules, dealing with the synthesis, analysis, perception, and manipulation of musical sound, in order to investigate how music can be incorporated into core curricula to inspire greater interest in math, physics, and computer science and to teach core STEM concepts to at-risk youth.
