Pan Xu Has Two Papers Accepted to Highly Competitive ICML Conference in Vienna, Austria
Assistant Professor Pan Xu in the Department of Computer Science has been honored with having two of his papers selected by the International Conference on Machine Learning (ICML), one of three primary conferences of high impact in machine learning and artificial intelligence. The co- authored paper on Promoting External and Internal Equities Under Ex-Ante/Ex-Post Metrics in Online Resource Allocation was one of only 3.5% accepted in the highly competitive Spotlight category. His single authored paper, Parameter-Dependent Competitive Analysis for Online Capacitated Coverage Maximization through Boostings and Attenuations, was also accepted as one of 27.5% in the Poster designation. Both papers were chosen from over 9,400 submissions.
According to Xu, there are many applications for machine learning (ML), but designing a system to work is more science than engineering. The first paper, which he co-authored with Karthik Abinav Sankararaman and Aravind Srinivasan from Meta and University of Maryland, College Park, respectively, aims to make algorithms perform better by what he terms “building a better brain for the system” so that the machine responds more intelligently in the manner of a human.
The paper proposes two models for better predicting future stochastic optimization that will improve equity and fairness in resource allocation. Xu envisions his solution as a means to address the uncertainty of the future by exploiting it.
Taking advantage of predictions and developing an allocation strategy in advance could have implications for everything from rationing limited physical resources to determining manual ones for efforts such as vaccine distribution based on internal factors like gender, race and age.
“For example, COVID vaccines were unequally distributed to populations who had better insurance over those who were underserved or disadvantaged. Better understanding internal populations can impact external equity,” Xu said, referencing an earlier study he conducted on designing equitable methods of distributing vital resources during widespread emergencies.
“A fire department knows that there will be accidents but not sure when or where they will occur or their size. This research aims to aid in immediate decision-making and committing remaining assets to where they are needed most.”
Xu’s second paper addresses the common occurrence of what is known as coverage maximization by designing a more efficient algorithm that adapts to special configurations of an instance while reducing cost. He offered the scenario of needing to implement a hotspot for a large event such as a convention. Rather than relying on a single large perimeter to serve the entire population, he proposes creating a solution using a series of smaller circled perimeters that will maximize coverage and reception and economize output, thus limiting financial and other resources.
Both papers were presented at the ICML conference July 21 – July 27, 2024, in Vienna, Austria.