Jason (jcreed) wrote,
Jason
jcreed

I got this talk announcement this morning. It sounds like the epitome of theadana-porn.
Goemans and Linial conjectured that negative type metrics (metrics that are squares of Euclidean metrics) embed into L_1 with constant distortion. Negative type metrics arise naturally as solutions of semidefinite relaxations for Sparsest Cut and Graph Partitioning problems. The GL-conjecture implies that the "integrality ratio" for the SDP-relaxation is bounded by a constant (which gives constant factor approximation algorithm). The recent breakthrough algorithm of [Arora Rao Vazirani] gave O(\sqrt{log n}) upper bound on the integrality ratio and they too conjectured a constant upper bound.
Tags: funny, words
Subscribe
  • Post a new comment

    Error

    Anonymous comments are disabled in this journal

    default userpic

    Your reply will be screened

    Your IP address will be recorded 

  • 5 comments