【计算机科学】【2019.03】双边匹配市场的深度学习

在这里插入图片描述

本文为美国哈佛大学（作者：Shira Li）的学士论文，共70页。

在双边匹配的经济学理论问题中，我们寻求在市场双方的代理之间建立匹配，每个代理都对另一方的代理有排序偏好。特别令人感兴趣的是稳定匹配，在这种情况下，没有一对代理相互愿意与对方匹配，而不是与其指定的伙伴匹配。匹配算法不仅在理论上相关，而且可以成功地解决现实世界中的难题，例如如何将医学生与住院医师计划进行最佳匹配。

本文第一部分介绍了双边匹配理论，重点介绍了寻找稳定匹配的方法、稳定匹配集的结构以及稳定匹配机制下的策略行为问题。第二部分介绍了一个新的深度学习框架，该框架对一对一匹配市场进行建模，并证明了其学习近似稳定匹配的能力，成功地复制了Gale-shapley的原始稳定性结果。此外，该框架还发现了既完整又能获得高总福利的匹配。到目前为止，人们对主导战略激励相容性、单个代理的战略问题和匹配的所有代理的总体稳定性之间的权衡还知之甚少。

In the economic theory problem of two-sided matching, we seek to creatematchings between agents on two sides of a market, each of whom has ranked,ordinal preferences over agents on the other side. Of particular interest are stable matchings,where no pair of agents would mutually prefer to be matched to each other than to their assigned partners. Not only theoretically relevant, matching algorithms can successfully solve real-world conundrums, such as how to optimally match medical students to hospitalresidency programs. The first part of this thesis gives an introduction to the theory of two-sided matching, focusing on methods of findingstable matchings, the structure of the set of stable matchings, and questions of strategic behavior understable matching mechanisms. The second partintroduces a novel deep learning framework that models one-to-one matchingmarkets and demonstrates its ability to learn approximately stable matchings,successfully replicating Gale-Shapleys original stability results. Furthermore, this framework finds matchings which are both complete and achieve high aggregate welfare. Given this, our framework holds promise forexamining the until now, poorly understood trade-offs between dominant-strategyincentive compatibility, a strategic question for individual agents, and theaggregate stability for all agents of a matching.

```
    历史回顾
```
```
    稳定匹配简介
```
```
    稳定匹配的结构
```
```
    对个体代理人的战略激励
```

    基于经济学理论的深度学习

```
    寻找近似的稳定匹配
```
```
    结论
```

下载地址：

https://url92.ctfile.com/f/1850492-516223241-6afe99

（访问密码：3660）

更多精彩文章请关注公众号：在这里插入图片描述