Like many, I fondly remember stories told by science communicators like Brian Greene and Neil deGrasse Tyson. They marveled at Newton’s laws then revealed the existence of more accurate theories whose existence had been unexpected: Special Relativity, General Relativity, Quantum Mechanics. The idea that nature could be misunderstood at a fundamental level, but that we could slowly, systematically reveal hidden layers led me to think of science as one of the most worthwhile pursuits.
Nonetheless, I became interested in computers. I knew scientific understanding had made possible their existence, but computers were interesting because I could design and implement programs – I was not motivated by a desire to discover things about machines. My interest led to degrees in Computer Science (engineering processes for machines) and Mathematics. Each was quite interesting on its own, but neither endeavor seemed to lead me to anything like the mystique of physics at the start of the 20th century. Mathematics led to unsolved problems which were all either grand and unsolvable (by me) or uninteresting. Computer Science seemed much the same. There were plenty of scientific problems, but none of them seemed that exciting. It was more worthwhile, more fun, to apply what I had learned to engineer software, like most of my peers.
So I was accepted as a co-op at IBM and found myself working on a data analytics project. This was a satisfying experience and left me with the impression that I could have a successful, fulfilling career in software development. The particular project I worked on left me with some knowledge of Machine Learning (ML) and a desire to pursue it further. Partly for the co-op, and partly for my own interest, I learned about the field and was lucky to discover how most of the tools I had pursued were used in this one area. That, by itself, wasn’t enough to maintain interest. Like ML, Numerical Analysis used my interests to a similar degree, but I had already ruled that out as uninteresting. Still, ML had a certain appeal, so I looked into it more.
While at IBM, I began taking ML related courses from Coursera including Geoff Hinton’s Neural Networks course. Soon after, I was accepted to do an MS in CS at VT and began working with Dr. Dhruv Batra in Machine Learning. Since I was already interested in Computer Vision, Dhruv seemed a good fit.
Recently, we’ve focused on a particular model called a Convolutional Neural Network (CNN), which helps illustrate a point I’ve discovered about Artificial Intelligence (AI) in general: AI is the next physics. Problems in this field sometimes feel grand – Relativity and Quantum Mechanics – and we’re making a lot of progress very quickly – physics in the first half of the 20th century. In 2012, some researchers edited the previous iteration of CNNs and ended up producing spectacular results on a fundamental problem in Computer Vision 1. That year and in every year since, large gains have been made on that and similar problems using CNNs 2. Progress like this and anticipation of harder problems yet to be solved, yet to be posed, are what drive me to participate in research.
Since my interests line up well and I draw motivation from the field’s potential, Artificial Intelligence and Machine Learning seem like the right place for me.