Learning, decision-making, and inference with limited experimentation
Abstract/Contents
- Abstract
- Over the last decade, machine learning techniques have revolutionized a variety of disciplines including medicine, natural language processing, computer vision, and finance. With this central role that machine learning has in our lives, there exists a vital need to fully investigate the applicability of these techniques, especially in human-centered domains. Indeed, many assumptions behind the theoretical analysis of machine learning algorithms are not satisfied in practice, which can lead to harmful consequences in sensitive applications. One example is in medical decision-making, where the impact of wrong decisions may be severe and irreversible. Furthermore, despite these huge advances in machine learning, the traditional randomized controlled trials are still the gold standard for testing the efficacy of various treatment and policy interventions. Although these trials are very robust and valid with minimal (or no) assumptions, they are very expensive and difficult to run, especially in medical applications. Thus, it is also important to carefully design and customize machine learning techniques that under a mild set of assumptions can provide meaningful and valid guarantees about the treatment effects using observational data. Observational data are usually more accessible and have the advantage of bypassing the burden of running experiments. This doctoral dissertation focuses on designing and analyzing data-driven methods with limited experimentation, focusing on three different settings. First, we study the online setting, where the decision-maker needs to personalize treatment decisions sequentially and wishes to reduce the amount of experimentation (randomization). In the rest of this dissertation, we consider the offline setting where the decision-maker has access to some observational data and wishes to estimate and draw inference about treatment effects. Specifically, we consider panel data models and discuss the treatment effect estimation using matrix completion methods. Moreover, we analyze personalized (non-parametric) inference from observational data with high dimensional covariates and combine non-parametric estimators with sub-sampling techniques to provide valid confidence intervals that are able to adapt to a priori unknown lower-dimensional structure of data.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2019; ©2019 |
| Publication date | 2019; 2019 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Khosravi, Khashayar |
|---|---|
| Degree supervisor | Bayati, Mohsen |
| Thesis advisor | Bayati, Mohsen |
| Thesis advisor | Johari, Ramesh, 1976- |
| Thesis advisor | Montanari, Andrea |
| Thesis advisor | Weissman, Tsachy |
| Degree committee member | Johari, Ramesh, 1976- |
| Degree committee member | Montanari, Andrea |
| Degree committee member | Weissman, Tsachy |
| Associated with | Stanford University, Department of Electrical Engineering. |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Khashayar Khosravi. |
|---|---|
| Note | Submitted to the Department of Electrical Engineering. |
| Thesis | Thesis Ph.D. Stanford University 2019. |
| Location | electronic resource |
Access conditions
- Copyright
- © 2019 by Khashayar Khosravi
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...