Cornell University
Advice-Augmented Algorithms for Online Matching and Resource Allocation
Abstract
dc:description.abstractReal life problems are full of uncertainty. How we handle it is important, since it affects the design and performance of algorithms. Often, the uncertainty is assumed to follow some known distribution, but in practice the estimate of the distribution may or may not be accurate. At other times, the uncertainty is assumed to be adversarial, but this can be too pessimistic for most real life instances. Advice-augmented algorithms aim to bridge the gap between these two models. In this framework, the algorithm is given some advice or prediction (e.g. from historical data, forecasts, or expert advice), whose quality is unknown. We aim to design algorithms that perform well when the quality is high (consistency), yet remain robust in their performance even when the quality is low (robustness). We consider advice-augmented algorithms for two problems. The first is two-stage matching: We design an algorithm that attains the optimal tradeoff between consistency and robustness. The second is Nash social welfare maximization in online resource allocation: We show that access to reasonable predictions gives an exponential improvement over the worst-case performance. Convex optimization plays a key role in both results.
Degree
thesis:*- Name thesis:degree_name
- Ph. D., Operations Research and Information Engineering
- Level thesis:degree_level
- Doctor of Philosophy
- Discipline thesis:degree_discipline
- Operations Research and Information Engineering
- Grantor
- Cornell University
- Year dc:date.issued
- 2024
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Jin, Billy Zhengxu
- Chair dc:contributor.chair
-
- Williamson, David
- Committee members dc:contributor.committeemember
-
- Shmoys, David
- Banerjee, Siddhartha
Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-NoDerivatives 4.0 International
- Licence dc:rights.uri
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Dc Identifier Bibid
- 16611787
- Dc Identifier Other
-
ProQuest Submission ID: 14568
ProQuest Publication ID: 31488353