Mathematics Colloquia and Seminars
Uniform Sampling through the Lovász Local LemmaMathematical Physics & Probability
|Speaker:||Jingcheng Liu, UC Berkeley|
|Start time:||Wed, Sep 26 2018, 3:10PM|
Abstract: We propose a new algorithmic framework, called “partial rejection sampling”, to draw samples exactly from a product distribution, conditioned on none of a number of bad events occurring. Our framework builds (perhaps surprising) new connections between the variable framework of the Lovász Local Lemma and some classical sampling algorithms such as the “cycle-popping” algorithm for rooted spanning trees by Wilson. Among other applications, we discover new algorithms to sample satisfying assignments of k-CNF formulas with bounded variable occurrences.
Joint work with Heng Guo and Mark Jerrum.