A professor of Computer Science, Dr. Devika Subramanian has been at Rice for 18 years. In her research, she develops data-driven prediction algorithms using artificial intelligence and machine learning and applies them to problems in engineering, natural sciences as well as the social sciences. Her algorithms have been used to predict mortality in patients with chronic heart failure, the risk of hurricane wind damage to individual homes in Harris County, as well as outbreak of conflict in the Middle East.
She teaches Computer Science 440, an upper level computer science course on artificial intelligence, which is defined as “the discipline of designing intelligent agents.” The students in this course come from various schools of studies and disciplines in engineering. The class has the usual set of programming assignments and problem sets, typical of any upper level computer science class. For these, Dr. Subramanian tells her students, “the weeds are cleared out and the path is pretty smooth. If you follow the steps in these assignments, you will reach the destination.” But there is more to the class.
She explains, “I also want to give them an unstructured problem-solving experience where I throw them in unfamiliar terrain, I tell them upfront, ‘Look, you will solve a brand-new problem, one that has no known solution. You will have to plot your own course through this terrain and find your way to the answer.’” In COMP 440, students work in two-person teams in a semester long project called Pacwar, a cellular-automata based game. The behavior of a Pacwar species is determined by a genome of length 50, each position in the genome can have one of four possible values. Species duel each other in a fixed size arena (shown below) and there are specific scoring rules for the duel. There are 4^50 possible species and the goal is to find a species that can defeat all the species created by their classmates from the present year as well as winners from the past 18 years. Running duels between all pairs of species would take longer than the universe has been in existence!
A screenshot of PacWar
Dr. Subramanian was introduced to Pacwar herself in the best class she has ever taken in her academic career. During her first year as a graduate student at Stanford, she took a class on problem solving with the computer science legend, Donald Knuth. Knuth handed out six unsolved problems at the start of the term, a variation of Pacwar was one of those problems. She was hooked, and she decided she would give her own students that intense experience in modeling and solving a computationally challenging problem.
Solving Pacwar requires a mix of human ingenuity and brute force computation. It highlights the power of the symbiosis of human and machine intelligence. The amount of compute power needed is beyond the capability of individual laptops. This is where the Office of Information Technology’s (OIT) Karl Burkett came in. Burkett, the Linux Systems Administrator for Education started collecting out-of-service servers and restoring them. With the addition of 15 repurposed, fully functioning servers, students could now run their species finding algorithms without overloading their personal computers or monopolizing machine cycles on other servers on the Rice campus. These servers were dedicated to the students in the course for the duration of the semester and are available, for a limited time, for use upon special request or need.
With the help of a committed teacher and a resourceful staff member, students were able to successfully implement their species design ideas and duel it out with each other at the end of the semester to see which Pacmite reigned supreme. Dr. Subramanian hopes that her students leave the class with the confidence to tackle any problem, “the confidence that they actually have the intellectual strategies, and the computational prowess to successfully deploy them.”