Reproduce “Exploring and Interacting with the Set of Good Sparse Generalized Additive Models” 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
Chudi Zhong, Zhi Chen, Jiachang Liu, Margo Seltzer, Cynthia Rudin
Researchers from Duke University and The University of British Columbia developed efficient methods to approximate the Rashomon set of sparse Generalized Additive Models (GAMs). This allows domain experts to explore, constrain, and directly modify near-optimal models while maintaining predictive performance and fostering greater trust in machine learning systems.
Key insights
- The Rashomon set for sparse GAMs can be effectively approximated by a convex set (an ellipsoid), which allows for efficient exploration and interaction with near-optimal models.
- Deriving approximations for numerous smaller support sets from a single large approximation using the blocking method is significantly faster (up to 1100x) and maintains high fidelity compared to repeated full optimizations.
- Interactive exploration of the Rashomon set enables robust variable importance analysis, helps identify the consistency of shape function patterns, and allows users to incorporate specific domain constraints or direct edits while preserving high model performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.