Deep Assortment Optimization

Zhen Yang, Jiayou Liang, Yang An, Zhi Wang, Rui Gao, Shuang Li

Keywords: Assortment optimization, Lovasz extension, Choice model

TL;DR: We develop a new differentiable stochastic optimization framework for assortment optimization based on Lovasz extension and randomized rounding with provable guarantees.

Abstract: Assortment optimization involves selecting a subset of items that maximizes expected reward under a choice model, with applications in online platforms and revenue management systems. We propose a differentiable, stochastic optimization framework that applies the Lovász extension to embed the discrete objective into the unit hypercube. The resulting continuous problem is solved efficiently with stochastic gradient descent and converted back to a near-optimal discrete assortment via a rounding scheme. Our method is scalable, model-agnostic, offers theoretical guarantees, and naturally extends to cardinality-constrained settings.

Primary Area: optimization

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.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2026/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 15644