AarXiv12 Jun 2023Cited 16×

Cross-Institutional Transfer Learning for Educational Models: Implications for Model Performance, Fairness, and Equity

J. Scott Gardner, Renzhe Yu, Quan Nguyen, Christopher Brooks, René F. Kizilcec

Abstract

'René F. Kizilcec'] Josh Gardner, Renzhe Yu, Quan Nguyen, Christopher Brooks, and Rene F. Kizilcec. 2023. Cross-Institutional Transfer Learning for Educational Models: Implications for Model Performance, Fairness, and Equity. In 2023 ACM Conference on Fairness, Accountability, and Transparency (FAccT '23), June 12–15, 2023, Chicago, IL, USA. ACM, New York, NY, USA, 21 pages. https: //doi.org/10.1145/3593013.3594107

A figure from Cross-Institutional Transfer Learning for Educational Models: Implications for Model Performance, Fairness, and Equity — fig. from the paper

§ The Valyu brief

The paper, in five minutes

Read from the full text via Valyu Contents

Zero-shot cross‑institutional transfer of student dropout models matches local performance and fairness across four universities, with ensembling offering no extra benefit.

What’s new

i.Proposed a framework and metrics for evaluating utility and fairness of cross‑institutional dropout prediction models.
ii.Conducted an empirical study on over 200,000 students from four universities without sharing raw training data.
iii.Showed that simple zero‑shot transfer matches locally‑trained model performance and fairness.
iv.Demonstrated that stacked ensembling provides no additional performance or fairness gains.
v.Highlighted the need for intersectional fairness analysis to uncover hidden disparities.

How they did it

The authors trained student dropout prediction models at individual institutions and evaluated zero‑shot transfer to other institutions, respecting data‑sharing constraints. They also tested stacked ensembling and measured utility and fairness using proposed metrics, including intersectional analyses.

What they found

Zero‑shot cross‑institutional transfer achieved similar performance to locally‑trained models for all four institutions.
Stacked ensembling did not improve overall performance or fairness compared to zero‑shot or local models.
No fairness‑accuracy tradeoff was observed across dozens of models and transfer schemes.
Study used data representing over 200,000 enrolled students annually.

§ The Valyu brief

Reading the full paper and taking notes. This takes a few seconds…

§ Ask this paper

Ask a question about this paper

Valyu reads the full text and answers from what the paper actually says.

§ Related work in other archives

Cross-Institutional Transfer Learning for Educational Models: Implications for Model Performance, Fairness, and Equity

The paper, in five minutes

Ask a question about this paper

Similar papers in all four archives

Similar papers in all four archives

The paper, in five minutes