前沿进展:在组合优化问题中重新审视采样
导语
![](/wp-content/uploads/2023/11/wxsync-2023-11-cb6bcab164110a8ed49c32f9502170d0.png)
目录
-
采样算法概述
-
高效的离散空间采样算法
-
采样算法与组合优化
-
采样算法与AI模型
1. 采样算法概述
1. 采样算法概述
![](/wp-content/uploads/2023/11/wxsync-2023-11-f393a25e9c239453e89e8e0bb11968d5.png)
![](/wp-content/uploads/2023/11/wxsync-2023-11-fb49a8aa41c18277e30b8fe82551cb5d.png)
![](/wp-content/uploads/2023/11/wxsync-2023-11-fb49a8aa41c18277e30b8fe82551cb5d.png)
![](/wp-content/uploads/2023/11/wxsync-2023-11-895bab458cada2b0da86e84358b99cb4.jpeg)
2. 高效的离散空间采样算法
2. 高效的离散空间采样算法
Sampling as first-order optimization over a space of probability measures [2022]. Anna Korba, Adil Salim. The variational formulation of the Fokker-Planck equation [1998]. Richard Jordan et al.
![](/wp-content/uploads/2023/11/wxsync-2023-11-007ddef9bcfd3d9f6e94c310e5360874.png)
![](/wp-content/uploads/2023/11/wxsync-2023-11-a44268fa39f67a4dd05fc2ef8c8c0ce4.png)
Fokker-Planck equations for a free energy functional or Markov process on a graph [2012]. Shui-Nee Chow et al.
Discrete Langevin sampler via Wasserstein gradient Flow [2023]. Haoran Sun et al. Thermodynamics of stoichiometric biochemical networks in living systems far from equilibrium [2005]. Hong Qian, Daniel A Beard. Path auxiliary proposal for MCMC in discrete space [2022]. Haoran Sun et al.
![](/wp-content/uploads/2023/11/wxsync-2023-11-019a8b02d0fa000693caa2db3f8323f8.png)
3. 采样算法与组合优化
3. 采样算法与组合优化
![](/wp-content/uploads/2023/11/wxsync-2023-11-cca82e16662375cdd985108ea1e23a83.png)
![](/wp-content/uploads/2023/11/wxsync-2023-11-873dbfb152d68f6dc61123f94f7e3cf7.png)
![](/wp-content/uploads/2023/11/wxsync-2023-11-3d9b8c54fb0a85551ddf168711392a6c.png)
![](/wp-content/uploads/2023/11/wxsync-2023-11-b53a91499a566127b98b2bfa04edcd6b.png)
![](/wp-content/uploads/2023/11/wxsync-2023-11-be70958dfab265e6a221aa60eb262398.png)
Revisiting Sampling for Combinatorial Optimization [2023]. Haoran Sun et al.
4. 采样算法与AI模型
4. 采样算法与AI模型
![](/wp-content/uploads/2023/11/wxsync-2023-11-5e43643c405f3a51574ac9777f5e8393.png)
![](/wp-content/uploads/2023/11/wxsync-2023-11-8c3883bc5b6f7aa8c5ac8cbb62edac7e.png)
![](/wp-content/uploads/2023/11/wxsync-2023-11-ce30240386163702962dcc8dd66b1f07.png)
学者简介
戴涵俊,Google DeepMind实验室资深科学家和研究经理,此前于佐治亚理工获得博士学位。研究方向:高效的生成模型,包括大语言模型,图像和结构化数据模型,及其采样和优化的底层算法
图神经网络与组合优化读书会
![](/wp-content/uploads/2023/11/wxsync-2023-11-6d3015426f4fff8640b130846d298423.png)
推荐阅读