PROGRAM
ODELJENJE ZA MATEMATIKU MATEMATIČKOG INSTITUTA SANU |
PROGRAM ZA SEPTEMBAR 2023.
PETAK, 29.09.2023. u 12:30, Kneza Mihaila 36, sala 301f i On-line
Dragan Stevanović, Matematički institut SANU
REINFORCEMENT LEARNING FOR GRAPH THEORY - CROSS-ENTROPY METHOD AND BASIC GRAPH ENVIRONMENTS
Adam Zsolt Wagner [arXiv:2104.14516] recently showed how reinforcement learning can be applied to construct (counter)examples in graph theory. We will showcase here a more readable, more stable and significantly faster reimplementation of his approach. We further implement several new reinforcement learning environments that cover constructions of simple graphs and trees, their signed variants, and graph with bounded maximum vertex degree. The presented concepts are illustrated by constructing counterexamples for several conjectures from literature.