Researchers Refute a Widespread Belief About Online Algorithms | Quanta Magazine

Algorithm designers try to craft online strategies that approach the offline ideal, whittling this ratio down toward 1.

In a paper first published online last November, three computer scientists showed that this isn’t always achievable.

The work offers “deeper insight into the central problem in this area’s central problem, Anupam Gupta.

Researchers used algorithms that include random decisions to rob the adversary of some of its power.

Researchers have suspected since the early 1990s that you can always find a randomized algorithm that reaches a specific performance goal: a competitive ratio proportional to log k.

This is called the randomized k-server conjecture, and researchers have shown that it’s true for some spaces, or specific collections of points. one ghtText__NxlGi">Quanta Quanta Rabani Marcin Bieńkowski ass="summaryFeed_highLightText the University of Wrocław pan Poland "summaryFeed_highLightText__NxlGi">Gupta the 2023 S the late 1980s eory of Computing twothe decades ass="summaryFeed_highLightText__NxlGi">“The Road Not Taken the Robert Frost Rabani one last November tT three NxlGi">two first two Anupam Gupta _NxlGi">log k. Carnegie Mellon University ghtText__NxlGi">first Coester Computer d_highLightT first NxlGi">Rabani 1988 ass="summaryFeed_highLightText__NxlGi">log k Coester the University of Oxford Christian Coester Microsoft Research Sébastien Bubeck Rabani log k the early 1990s the Hebrew University of Jerusalem Yuval Rabani one Bieńkowski two 1 ghtText__NxlGi">one just two Text__NxlGi">two one first