Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.
I love personalization. But sometimes one size might fit all. And learning such policies can need extra data & be more costly to deploy. In Science https://www.science.org/doi/10.1126/science.aeb9506 Zhaoqi Li & I share a statistical estimator of the benefit of personalization over a universal intervention
This is a simple test for personalization. They learn a personalized policy and best constant action on one subset. Then learn outcome and propensity models on another subset. Then evaluate an AIPW score on held-out data, followed by cross-fitting. https://arxiv.org/abs/2607.08951
Our k-fold personalization test takes in a dataset and has valid type 1 error and under certain assumptions, achieves semiparametric efficiency. We demonstrate its use across applications including student online course completion, depression treatment, and joke recommendations.
There are essential details about cross-fitting that I couldn't manage to squeeze in the character limit. But the process is about estimating the value gained by personalization, and proposing an associated test. PS: One of the authors (Brunskill) has long been interested in
I love personalization. But sometimes one size might fit all. And learning such policies can need extra data & be more costly to deploy. In Science https://www.science.org/doi/10.1126/science.aeb9506 Zhaoqi Li & I share a statistical estimator of the benefit of personalization over a universal intervention
This is a simple test for personalization. They learn a personalized policy and best constant action on one subset. Then learn outcome and propensity models on another subset. Then evaluate an AIPW score on held-out data, followed by cross-fitting. https://arxiv.org/abs/2607.08951
Our k-fold personalization test takes in a dataset and has valid type 1 error and under certain assumptions, achieves semiparametric efficiency. We demonstrate its use across applications including student online course completion, depression treatment, and joke recommendations.
There are essential details about cross-fitting that I couldn't manage to squeeze in the character limit. But the process is about estimating the value gained by personalization, and proposing an associated test. PS: One of the authors (Brunskill) has long been interested in
We release code for our K-Fold personalization estimator here: https://cs.stanford.edu/people/ebrun/kpe.html and the arxiv version is here: https://arxiv.org/abs/2607.08951
educational data mining / intelligent tutoring, and on looking at the paper, testing for personalization in an educational context is a clear motivation.
Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.
educational data mining / intelligent tutoring, and on looking at the paper, testing for personalization in an educational context is a clear motivation.