Reproduce “Neural-Bayesian Structure Learning for Discrete Choice Modeling” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Hyunsoo Yun, Eun Hak Lee, Jiaru Zhang, Ziran Wang, Eui-Jin Kim

Neural-Bayesian Structure Learning (Neural-BSL) integrates data-driven structure learning over traveler attributes with random utility-based discrete choice modeling. This framework allows for joint estimation of attribute dependencies and choice behavior, enabling policy simulations to account for cascading effects of interventions, and yields competitive predictive performance while providing interpretable causal insights.

Key insights

  • Jointly learning attribute dependencies and choice behavior within a single differentiable framework allows the discovered structural relationships to directly inform and enhance the discrete choice model's parameters.
  • Preserving random utility theory by treating choice as an outcome of utility comparison, rather than a structural node in the DAG, leads to more coherent and behaviorally sound structural insights for traveler attributes.
  • The ability to propagate interventions along the learned DAG provides a more realistic and mechanistically transparent approach to policy simulation, predicting both mode-share changes and the accompanying shifts in downstream attributes.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.