ECE Researchers win Outstanding Paper Award from ICML
ECE researchers have been selected as one of the six winners of the ICML 2023 Outstanding Paper Awards.
August 18, 2023 Isabella Mormando
The International Conference on Machine Learning (ICML) has selected a paper published by ECE professor, Do Young Eun, and his Ph.D. student, Jie Hu, as one of the six winners of its 2023 Outstanding Paper Awards. The paper was also authored by a recent Operation Research alumnus, Vishwarag Doshi, who was a former Ph.D. student of Eun and graduated in 2022.
There were 6,538 papers submitted to the ICML and only 1,827 papers were accepted for the competition. This puts our researchers’ paper in the top 0.1 percent. In order to filter down the best papers, the process involved selecting papers with high average scores, as well as those recommended by members of the program committee for consideration. The committee considered these papers and selected the award papers due to their excellent clarity, insight, creativity, and potential for lasting impact.
The paper is titled “Self-Repellent Random Walks on General Graphs – Achieving Minimal Sampling Variance via Nonlinear Markov Chains.”
Eun described this paper by saying “imagine a random walker on a graph, where the walker gets to choose one of the neighbors to move into, constrained by the local connectivity of the graph. This is basically the way a Markov chain operates, and forms a foundation to MCMC (Markov Chain Monte Carlo) — a fundamental component in randomized algorithms, graph sampling, stochastic optimization, statistical inference, and distributed machine learning, to list a few, as a means to attain a given distribution over the graph with local information.
Our work shows that there is a simple way to greatly enhance the performance by incorporating the entire past history into the walker’s next step — in a `self-repellent’ manner, thus making it sort of “nonlinear Markov” as opposed to being the usual (linear) Markov. What’s surprising is that this nonlinear Markov chain can produce samples with `near-zero’ variance (asymptotically) of any estimator driven by such a self-repellent walker. This means that such self-repellent walks can outperform an `imaginary’ random jumper/flier who can instantaneously jump to any node of the graph in one step, while still walking on the graph. In other words, our work hints that a walker can be made better than a random jumper!”
The ICML is the premier gathering of professionals dedicated to advancing the branch of artificial intelligence known as machine learning.
ICML is globally renowned for presenting and publishing cutting-edge research on all aspects of machine learning used in closely related areas like artificial intelligence, statistics and data science, as well as important application areas such as machine vision, computational biology, speech recognition and robotics.
ICML is one of the fastest-growing artificial intelligence conferences in the world. Participants at ICML span various backgrounds, from academic and industrial researchers to entrepreneurs and engineers to graduate students and postdocs.
“It’s unbelievable. Considering the sheer number of submitted papers and a wide range of areas covered under machine learning, we still feel lucky for our work to be selected as one of only 6 outstanding paper awards in this prestigious ICML, for which we are very proud of,” says Eun.
Congratulations to our ECE researchers on their outstanding recognition from the ICML!