George Montañez, a computer science faculty member at Harvey Mudd College, has won the IEEE SMC Award for best student paper at the 2017 IEEE Systems, Man, and Cybernetics Society Conference.
Montañez presented the paper “The Famine of Forte: Few Search Problems Greatly Favor Your Algorithm” to conference attendees this month in Banff, Canada.
“In the same spirit of abstractions in physics that allow us to talk about magnetism and electricity at the same time, or energy being interchangeable with matter, this paper introduces a general abstraction of machine learning as a type of search that allows one to simultaneously reason about many areas of machine learning,” says Montañez, who submitted the paper while still a PhD student at Carnegie Mellon University.
Using the abstraction, he proves results showing that for any fixed algorithm, most search problems are not greatly favorable, and for any fixed search problem, most algorithm strategies are not favorable. Therefore, matching problems to algorithms is provably difficult, no matter which is held fixed and which varies.
“Of practical importance, if an algorithm greatly excels on one class of problems, that class of problems must be relatively small,” he says. “If we further assume a restriction to only that class of problems for which an algorithm can do well, I show how dependence within that class between data resources and search targets bounds how likely that algorithm is to succeed. The paper then draws inferences based on these results, ruling out such things as ‘one-size-fits-all’ fitness functions in genetic algorithms and explains the modern trend to develop flexible learning methods like deep neural networks.”
Montañez’s work was chosen from over 1,200 submitted papers. “The paper is one that I have been refining for some time, so I was very happy to receive this honor for what I consider to be my best work,” he says.
This latest honor is one of a string of recent best paper awards for Montañez. At the 2017 International Joint Conference on Neural Networks, Montañez won the Best Poster award and the INNS/Intel Best Student Paper award for “The LICORS Cabinet: Nonparametric Light Cone Methods for Spatio-Temporal Modeling,” coauthored with Cosma Rohilla Shalizi. He also won the Best Paper award at CIKM 2014, with coauthors Ryen White and Xiao Huang for their paper on cross-device search.
Montañez, whose research lies at the intersection of computer science, algorithmic search and mathematics, is working at Microsoft and will join the Harvey Mudd faculty in fall 2018. He holds a PhD in machine learning from Carnegie Mellon University, an M.S. in computer science from Baylor University and a B.S. in computer science from the University of California Riverside. He was an NSF Graduate Research Fellow and a Ford Foundation Predoctoral Fellow and served as an intern at Microsoft Research and Yahoo! Labs. He has worked as a software engineer and full-stack developer.