BriefGPT.xyz
Nov, 2023
基于图的组合优化的变分退火
Variational Annealing on Graphs for Combinatorial Optimization
HTML
PDF
Sebastian Sanokowski, Wilhelm Berghammer, Sepp Hochreiter, Sebastian Lehner
TL;DR
通过引入子图标记化技术，我们在无监督学习和组合优化问题中实现了优越性能。使用自回归方法捕捉解变量之间的统计依赖关系，理论上证明并实证了热退火熵正则化的重要性以实现高效和稳定的学习。
Abstract
Several recent
unsupervised learning
methods use probabilistic approaches to solve
combinatorial optimization
(CO) problems based on the assumption of statistically independent solution variables. We demonstrate
→