Computational Creativity

Creativity plays a key role in many aspects of (intelligent) behavior, including:

  • Problem solving
  • Scientific Discovery
  • Art (visual metaphor)
  • Music
  • Language (metaphor, narrative, poetry, humor)
  • Design
Can we build computational systems that produce interesting/useful results through what must be attributable as creative means? If so, what does this mean? If not, why? Can these questions even be answered? Our research focuses on building systems to which we can attribute creativity, thus expanding our understanding of (artificial) intelligence and leading to the development of more robust artificially intelligent systems. In addition, work on computational creativity necessitates advances in other fields, such as natural language processing and understanding, computer vision, search and information retrieval. One example is an artificial artist, DARCI, that is learning to appreciate and produce visual art. DARCI recently acted as the sole juror for a art show called Fitness Function. We are also developing systems for automatic music composition that include the ability to incorporate emotional content and to produce music from non-musical inspirational sources and even for recipe generation.

Medical Applications of Artificial Intelligence

Modern medical procedure provides vast amounts of patient data that is grossly under-utilized, yet has the potential to revolutionize the diagnosis and treatment of many serious medical conditions. According to the Centers for Disease Control and Prevention, heart disease is the leading cause of death in the United states, attributable to more than 650,000 deaths each year (about 27\% of all U.S. deaths). Pulmonary artery pressure is an indicator of some types of heart and blood diseases, including pulmonary hypertension and sickle-cell anemia. Right-heart catheterization is the most accurate method for estimating pulmonary artery pressure; however, it is also an invasive procedure that is expensive, exposes patients to the risk of infection, and is not suited for long-term monitoring situations.

The medical literature points to a correlation between heart sound and pulmonary artery pressure. The goal of this work is to develop a prototype system that estimates a patient’s pulmonary artery pressure by analyzing heart sounds from the patient. The challenge is two-fold: 1) discover heart sound feature sets that will be useful as input to the system, and 2) build machine learning algorithms using those features that can act as pulmonary artery pressure estimators. We are currently working with a small cohort of patient data, to build diagnostic models (both classification- and regression-based). This work will naturally lead to the study of additional medical applications to which machine learning may be applied.

Traditional Machine Learning and Artificial Intelligence

Machine Learning and Artificial Intelligence are broad fields and our interests and past projects are diverse, including:

  • Theoretical and applied results in spectral learning
  • Price prediction for eBay markets
  • Algorithms for training recurrent spiking neural networks
  • Bayesian algorithms for computer vision
  • Multi-agent learning
  • Cognitive modeling for behavioral animation
  • Content-based music retrieval