[Theory] NOW: 2/27 TTIC Colloquium: Rediet Abebe, Harvard University
Mary Marre
mmarre at ttic.edu
Tue Feb 27 10:51:08 CST 2024
[image: Revised flyer.jpg]
*TALK DETAILS*
*When:* Tuesday, February 27, 2024 at* 11:00** a**m CT *
*Where: *Talk will be given *live, in-person* at
TTIC, 6045 S. Kenwood Avenue
5th Floor, Room 530
*Virtually:* *via *Panopto (*livestream
<https://uchicago.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=1a6f02ca-5a65-4d96-a787-b11e01877c17>*
)
* *limited access: see info below*
*Who: * Rediet Abebe, Harvard University
------------------------------
*Title: *When Does Allocation Require Prediction?
*Abstract:* Algorithmic predictions are emerging as a promising solution
concept for efficiently allocating scarce societal resources. Fueling their
use is an underlying assumption that predictive systems are necessary for
*identification---*that we can target resources more efficiently through
individual risk scores output by such systems. We examine this assumption
empirically and theoretically.
Empirically, we present findings from a large-scale evaluation of
Wisconsin's Dropout Early Warning System (DEWS)---an early warning system
used to predict each public school student's likelihood of dropping out of
high school. Using nearly a decade's worth of data, we show that DEWS
accurately sorts students by their dropout risk, and it may have resulted
in a single-digit increase in graduation rates. However, a simple
allocation mechanism that only uses environmental information about
schools, neighborhoods, and districts may have sufficed for targeting
interventions just as efficiently.
We examine this gap that emerges between predictions and allocations
theoretically. Using a simple mathematical model, we evaluate the efficacy
of individual prediction-based allocation mechanisms with
environmentally-based mechanisms, which only use aggregate school-level
statistics. We find that individual prediction-based allocations outperform
environmental-based allocations only when between-school inequality is low
or the allocation budget is high. Our theoretical findings hold for a wide
range of settings for the heterogeneity of treatment effects, learnability
of school-level statistics, and price of prediction.
These insights call into question the necessity of individual predictions
for efficient allocations when outcomes are structurally determined.
Predictions may only improve allocations only if inequality is low.
*This talk is based on joint work with Tolani Britton, Moritz Hardt, Juan
Carlos Perdomo, and Ali Shirali. It is informed by discussions with
Wisconsin's Department of Public Instruction, particularly Erin Fath, Carl
Frederick, and Justin Meyer. *
*Bio: *Rediet Abebe <https://www.redietabebe.com/> is a Junior Fellow at
the Harvard Society of Fellows <https://socfell.fas.harvard.edu/> and an Andrew
Carnegie Fellow
<https://www.carnegie.org/awards/andrew-carnegie-fellows/2022/>. Abebe’s
research examines the interaction of algorithms and inequality, with a
focus on contributing to the scientific foundations of this emerging
research area. Abebe co-launched the ACM Conference on Equity and Access in
Algorithms, Mechanisms, and Optimization (ACM EAAMO <https://eaamo.org/>),
for which Abebe serves on the executive committee and was a program
co-chair for the inaugural conference. Abebe’s work has received
recognitions including the MIT Technology Reviews’ 35 Innovators Under 35,
the Bloomberg 50 as a one to watch, the ACM SIGKDD Dissertation Award, and
an honorable mention for the ACM SIGecom Dissertation Award.
*Host: **Avrim Blum* <avrim at ttic.edu>
*Access to this livestream is limited to *TTIC / UChicago* (press panopto
link and sign in to your UChicago account with CNetID).
Mary C. Marre
Faculty Administrative Support
*Toyota Technological Institute*
*6045 S. Kenwood Avenue, Rm 517*
*Chicago, IL 60637*
*773-834-1757*
*mmarre at ttic.edu <mmarre at ttic.edu>*
On Tue, Feb 27, 2024 at 8:30 AM Mary Marre <mmarre at ttic.edu> wrote:
> [image: Revised flyer.jpg]
>
> *TALK DETAILS*
>
> *When:* Tuesday, February 27, 2024 at* 11:00** a**m CT *
>
>
> *Where: *Talk will be given *live, in-person* at
>
> TTIC, 6045 S. Kenwood Avenue
>
> 5th Floor, Room 530
>
>
> *Virtually:* *via *Panopto (*livestream
> <https://uchicago.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=1a6f02ca-5a65-4d96-a787-b11e01877c17>*
> )
>
> * *limited access: see info below*
>
>
> *Who: * Rediet Abebe, Harvard University
> ------------------------------
> *Title: *When Does Allocation Require Prediction?
>
> *Abstract:* Algorithmic predictions are emerging as a promising solution
> concept for efficiently allocating scarce societal resources. Fueling their
> use is an underlying assumption that predictive systems are necessary for
> *identification---*that we can target resources more efficiently through
> individual risk scores output by such systems. We examine this assumption
> empirically and theoretically.
>
>
> Empirically, we present findings from a large-scale evaluation of
> Wisconsin's Dropout Early Warning System (DEWS)---an early warning system
> used to predict each public school student's likelihood of dropping out of
> high school. Using nearly a decade's worth of data, we show that DEWS
> accurately sorts students by their dropout risk, and it may have resulted
> in a single-digit increase in graduation rates. However, a simple
> allocation mechanism that only uses environmental information about
> schools, neighborhoods, and districts may have sufficed for targeting
> interventions just as efficiently.
>
>
> We examine this gap that emerges between predictions and allocations
> theoretically. Using a simple mathematical model, we evaluate the efficacy
> of individual prediction-based allocation mechanisms with
> environmentally-based mechanisms, which only use aggregate school-level
> statistics. We find that individual prediction-based allocations outperform
> environmental-based allocations only when between-school inequality is low
> or the allocation budget is high. Our theoretical findings hold for a wide
> range of settings for the heterogeneity of treatment effects, learnability
> of school-level statistics, and price of prediction.
>
>
> These insights call into question the necessity of individual predictions
> for efficient allocations when outcomes are structurally determined.
> Predictions may only improve allocations only if inequality is low.
>
>
> *This talk is based on joint work with Tolani Britton, Moritz Hardt, Juan
> Carlos Perdomo, and Ali Shirali. It is informed by discussions with
> Wisconsin's Department of Public Instruction, particularly Erin Fath, Carl
> Frederick, and Justin Meyer. *
>
>
> *Bio: *Rediet Abebe <https://www.redietabebe.com/> is a Junior Fellow at
> the Harvard Society of Fellows <https://socfell.fas.harvard.edu/> and an Andrew
> Carnegie Fellow
> <https://www.carnegie.org/awards/andrew-carnegie-fellows/2022/>. Abebe’s
> research examines the interaction of algorithms and inequality, with a
> focus on contributing to the scientific foundations of this emerging
> research area. Abebe co-launched the ACM Conference on Equity and Access in
> Algorithms, Mechanisms, and Optimization (ACM EAAMO <https://eaamo.org/>),
> for which Abebe serves on the executive committee and was a program
> co-chair for the inaugural conference. Abebe’s work has received
> recognitions including the MIT Technology Reviews’ 35 Innovators Under 35,
> the Bloomberg 50 as a one to watch, the ACM SIGKDD Dissertation Award, and
> an honorable mention for the ACM SIGecom Dissertation Award.
>
> *Host: **Avrim Blum* <avrim at ttic.edu>
>
> *Access to this livestream is limited to *TTIC / UChicago* (press panopto
> link and sign in to your UChicago account with CNetID).
>
>
>
>
> Mary C. Marre
> Faculty Administrative Support
> *Toyota Technological Institute*
> *6045 S. Kenwood Avenue, Rm 517*
> *Chicago, IL 60637*
> *773-834-1757*
> *mmarre at ttic.edu <mmarre at ttic.edu>*
>
>
> On Mon, Feb 26, 2024 at 5:13 PM Mary Marre <mmarre at ttic.edu> wrote:
>
>> [image: image.png]
>> *TALK DETAILS*
>>
>> *When:* Tuesday, February 27, 2024 at* 11:00** a**m CT *
>>
>>
>> *Where: *Talk will be given *live, in-person* at
>>
>> TTIC, 6045 S. Kenwood Avenue
>>
>> 5th Floor, Room 530
>>
>>
>> *Virtually:* *via *Panopto (*livestream
>> <https://uchicago.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=1a6f02ca-5a65-4d96-a787-b11e01877c17>*
>> )
>>
>> * *limited access: see info below*
>>
>>
>> *Who: * Rediet Abebe, Harvard University
>> ------------------------------
>> *Title: *When Does Allocation Require Prediction?
>>
>> *Abstract:* Algorithmic predictions are emerging as a promising solution
>> concept for efficiently allocating scarce societal resources. Fueling their
>> use is an underlying assumption that predictive systems are necessary for
>> *identification---*that we can target resources more efficiently through
>> individual risk scores output by such systems. We examine this assumption
>> empirically and theoretically.
>>
>>
>> Empirically, we present findings from a large-scale evaluation of
>> Wisconsin's Dropout Early Warning System (DEWS)---an early warning system
>> used to predict each public school student's likelihood of dropping out of
>> high school. Using nearly a decade's worth of data, we show that DEWS
>> accurately sorts students by their dropout risk, and it may have resulted
>> in a single-digit increase in graduation rates. However, a simple
>> allocation mechanism that only uses environmental information about
>> schools, neighborhoods, and districts may have sufficed for targeting
>> interventions just as efficiently.
>>
>>
>> We examine this gap that emerges between predictions and allocations
>> theoretically. Using a simple mathematical model, we evaluate the efficacy
>> of individual prediction-based allocation mechanisms with
>> environmentally-based mechanisms, which only use aggregate school-level
>> statistics. We find that individual prediction-based allocations outperform
>> environmental-based allocations only when between-school inequality is low
>> or the allocation budget is high. Our theoretical findings hold for a wide
>> range of settings for the heterogeneity of treatment effects, learnability
>> of school-level statistics, and price of prediction.
>>
>>
>> These insights call into question the necessity of individual predictions
>> for efficient allocations when outcomes are structurally determined.
>> Predictions may only improve allocations only if inequality is low.
>>
>>
>> *This talk is based on joint work with Tolani Britton, Moritz Hardt, Juan
>> Carlos Perdomo, and Ali Shirali. It is informed by discussions with
>> Wisconsin's Department of Public Instruction, particularly Erin Fath, Carl
>> Frederick, and Justin Meyer. *
>>
>>
>> *Bio: *Rediet Abebe <https://www.redietabebe.com/> is a Junior Fellow at
>> the Harvard Society of Fellows <https://socfell.fas.harvard.edu/> and an Andrew
>> Carnegie Fellow
>> <https://www.carnegie.org/awards/andrew-carnegie-fellows/2022/>. Abebe’s
>> research examines the interaction of algorithms and inequality, with a
>> focus on contributing to the scientific foundations of this emerging
>> research area. Abebe co-launched the ACM Conference on Equity and Access in
>> Algorithms, Mechanisms, and Optimization (ACM EAAMO <https://eaamo.org/>),
>> for which Abebe serves on the executive committee and was a program
>> co-chair for the inaugural conference. Abebe’s work has received
>> recognitions including the MIT Technology Reviews’ 35 Innovators Under 35,
>> the Bloomberg 50 as a one to watch, the ACM SIGKDD Dissertation Award, and
>> an honorable mention for the ACM SIGecom Dissertation Award.
>>
>> *Host: **Avrim Blum* <avrim at ttic.edu>
>>
>> *Access to this livestream is limited to TTIC / UChicago (press panopto
>> link and sign in to your UChicago account with CNetID).
>>
>>
>>
>>
>>
>>
>>
>>
>> Mary C. Marre
>> Faculty Administrative Support
>> *Toyota Technological Institute*
>> *6045 S. Kenwood Avenue, Rm 517*
>> *Chicago, IL 60637*
>> *773-834-1757*
>> *mmarre at ttic.edu <mmarre at ttic.edu>*
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/theory/attachments/20240227/be6efbc1/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 636324 bytes
Desc: not available
URL: <http://mailman.cs.uchicago.edu/pipermail/theory/attachments/20240227/be6efbc1/attachment-0001.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Revised flyer.jpg
Type: image/jpeg
Size: 340178 bytes
Desc: not available
URL: <http://mailman.cs.uchicago.edu/pipermail/theory/attachments/20240227/be6efbc1/attachment-0001.jpg>
More information about the Theory
mailing list