Multiple-Instance Feature Ranking
Andrew C Latham · OhioLink ETD Center (Ohio Library and Information Network) · 2016
I'd like to begin by thanking my advisor, Dr. Soumya Ray, who has provided excellent guidance over the last two years and played a critical role in developing not just my understanding of machine learning but also my ability to think critically, overcome obstacles, and reason about research topics in a way that I never would have been able to do on my own.The work I have done with Professor Ray has been incredibly rewarding and inspired a passion for learning and exploring that will stay with me long after I leave this institution.I would also like to thank Gary Doran, with whom I worked during my first year and a half of research.When I first started studying these topics, Gary was always happy to explain complicated ideas in an intuitive fashion, and had seemingly infinite patience when introducing me to the theoretical pillars that would later form the foundation for my own work.Gary also built the experimental framework that I used to execute the experiments in this thesis.In addition to Professor Ray, my committee is composed of Dr. Harold Connamacher and Dr. Michael Lewicki.In my time at Case Western Reserve University, the classes I took from these professors have proven to be among the most useful both in general and, in particular, in the writing of this thesis; from Professor Connamacher I learned the basics of mathematical reasoning in the context of computer science, and from Professor Lewicki I learned how to approach machine learning from a probabilistic perspective.Professor Connamacher also took the time to send me a vii ACKNOWLEDGMENTS detailed critique of this thesis prior to its submission, which has helped me to clarify and improve the explanations of many key concepts.I would like to thank all my friends at Case Western Reserve University, but especially Larry Muhlstein, Ben Cowen, and Greg Penzias, who have proven to be great sources of advice and camaraderie.In the final week before completion of this thesis, Greg allowed me to use his allocated server time to run my experiments, which allowed me to improve the experimental results section.Finally and most importantly, I would like to