Bayesian Preference Elicitation for Personalized Prefactual Recommendation
Primary Area: general machine learning (i.e., none of the above)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Bayesian Preference Elicitation, Recourse Generation, Mutual Information
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: A prefactual recommendation, also known as an algorithmic recourse, provides actionable guidance to an individual to overturn a machine learning prediction at a minimal cost of efforts. Existing methods impose an explicit assumption on the cost function, but in reality, different individuals may possess diverse and unique cost preferences. Failing to adapt the guidance to an individual's cost preference can lead to irrelevant and inefficient recommendations. To personalize the guidance to the individual cost, we propose a Bayesian preference elicitation framework that learns the cost function from the individual's feedback on a small number of pairwise comparisons. This framework relies on a sequential, mutual-information-maximization question-answering scheme to obtain a posterior distribution of an individual's cost weighting matrix. We then deploy this posterior to recommend a graph-based sequential guidance with minimal expected cost, leading the individual to achieve the desired algorithmic outcome. Numerical experiments on synthetic and real-world datasets demonstrate the power of our method in capturing the individual's preference and recommending personalized recourse.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
Supplementary Material: zip
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 4706
Loading