Consistent Estimators for Learning to Defer to an Expert
Learning algorithms are often used in conjunction with expert decision makers in practical scenarios, however, this fact is largely ignored when designing these algorithms. In this thesis, we explore how to learn predictors that can either predict or choose to defer the decision to a downstream expe...
Main Author: | |
---|---|
Other Authors: | |
Format: | Thesis |
Published: |
Massachusetts Institute of Technology
2023
|
Online Access: | https://hdl.handle.net/1721.1/151827 |
_version_ | 1811072958341316608 |
---|---|
author | Mozannar, Hussein |
author2 | Sontag, David |
author_facet | Sontag, David Mozannar, Hussein |
author_sort | Mozannar, Hussein |
collection | MIT |
description | Learning algorithms are often used in conjunction with expert decision makers in practical scenarios, however, this fact is largely ignored when designing these algorithms. In this thesis, we explore how to learn predictors that can either predict or choose to defer the decision to a downstream expert. Given only samples of the expert's decisions, we give a procedure based on learning a classifier and a rejector and analyze it theoretically. Our approach is based on a novel reduction to cost sensitive learning where we give a consistent surrogate loss for cost sensitive learning that generalizes the cross entropy loss. We show the effectiveness of our approach on a variety of experimental tasks. |
first_indexed | 2024-09-23T09:24:31Z |
format | Thesis |
id | mit-1721.1/151827 |
institution | Massachusetts Institute of Technology |
last_indexed | 2024-09-23T09:24:31Z |
publishDate | 2023 |
publisher | Massachusetts Institute of Technology |
record_format | dspace |
spelling | mit-1721.1/1518272023-08-24T03:49:19Z Consistent Estimators for Learning to Defer to an Expert Mozannar, Hussein Sontag, David Massachusetts Institute of Technology. Institute for Data, Systems, and Society Learning algorithms are often used in conjunction with expert decision makers in practical scenarios, however, this fact is largely ignored when designing these algorithms. In this thesis, we explore how to learn predictors that can either predict or choose to defer the decision to a downstream expert. Given only samples of the expert's decisions, we give a procedure based on learning a classifier and a rejector and analyze it theoretically. Our approach is based on a novel reduction to cost sensitive learning where we give a consistent surrogate loss for cost sensitive learning that generalizes the cross entropy loss. We show the effectiveness of our approach on a variety of experimental tasks. S.M. 2023-08-23T16:11:39Z 2023-08-23T16:11:39Z 2023-06 2023-07-24T18:01:48.706Z Thesis https://hdl.handle.net/1721.1/151827 Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) Copyright retained by author(s) https://creativecommons.org/licenses/by-sa/4.0/ application/pdf Massachusetts Institute of Technology |
spellingShingle | Mozannar, Hussein Consistent Estimators for Learning to Defer to an Expert |
title | Consistent Estimators for Learning to Defer to an Expert |
title_full | Consistent Estimators for Learning to Defer to an Expert |
title_fullStr | Consistent Estimators for Learning to Defer to an Expert |
title_full_unstemmed | Consistent Estimators for Learning to Defer to an Expert |
title_short | Consistent Estimators for Learning to Defer to an Expert |
title_sort | consistent estimators for learning to defer to an expert |
url | https://hdl.handle.net/1721.1/151827 |
work_keys_str_mv | AT mozannarhussein consistentestimatorsforlearningtodefertoanexpert |