Reliability and stability in statistical and machine learning problems
Abstract/Contents
- Abstract
- The increasing application of statistical machine learning techniques has substantial promise in diverse fields such as medicine, policy-making, and finance. Hence, very complex machine learning models are being built for use in automated decision-making protocols. However, with the growing sophistication of these models, they are often impossible to analyze statistically, and hence, the predictions made by such models often come with no guarantees. Furthermore, we often have data from heterogeneous sources, shifting distributions, or only have access to noisy or weakly supervised data (e.g., it may be expensive to obtain properly labeled data for a classification task). With such additional challenges, how can we trust the results we achieve when we apply machine-learned methods particularly in critical areas? Further, a shift in data generating distributions makes it challenging for statistical knowledge to generalize well. For instance, the performance of predictive models may drastically deteriorate when deployed on a new test set or a statistical finding from an experiment may not replicate in future experiments. Is it possible to identify situations where models or statistical findings are sensitive to changes in the data generating distributions? Can we suggest methods to transfer statistical knowledge more accurately in such situations? To address the above challenges, this dissertation focuses on developing methods that leverage conformal inference, distributional robustness, and causal inference. The efficacy of these methods are further demonstrated via extensive experiments including applications to real-world machine learning and statistical problems.
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 | 2022; ©2022 |
Publication date | 2022; 2022 |
Issuance | monographic |
Language | English |
Creators/Contributors
Author | Gupta, Suyash |
---|---|
Degree supervisor | Duchi, John |
Thesis advisor | Duchi, John |
Thesis advisor | Candès, Emmanuel J. (Emmanuel Jean) |
Thesis advisor | Donoho, David Leigh |
Thesis advisor | Rothenhaeusler, Dominik |
Degree committee member | Candès, Emmanuel J. (Emmanuel Jean) |
Degree committee member | Donoho, David Leigh |
Degree committee member | Rothenhaeusler, Dominik |
Associated with | Stanford University, Department of Statistics |
Subjects
Genre | Theses |
---|---|
Genre | Text |
Bibliographic information
Statement of responsibility | Suyash Gupta. |
---|---|
Note | Submitted to the Department of Statistics. |
Thesis | Thesis Ph.D. Stanford University 2022. |
Location | https://purl.stanford.edu/br983xt6266 |
Access conditions
- Copyright
- © 2022 by Suyash Gupta
- 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...