Research map / Architectures and efficiency / RNNs and normalization
Neural Basis Models for Interpretability
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace per-feature networks (NAM) with shared basis functions (NBM) | NAM · CA Housing, FICO, CoverType, CUB, iNat Birds, Common Objects | accuracy/AUROC/mAP: iNat acc@1 0.4194 -> 0.4227; Common Objects mAP 0.2056 -> 0.2168 | Helped |
| replace NAM with NBM on CA Housing | NAM · CA Housing | RMSE: 0.5721 -> 0.5638 (lower is better) | Hurt |
| use shared bases instead of per-feature MLPs | NAM · feature count > 10 | number of parameters: 5x-50x fewer parameters | Helped |
| use shared bases with sparse optimization | NAM · Newsgroups | throughput: 23 -> 9K x/sec (x391) | Helped |
| NBM versus NAM on shape-function stability | NAM · CA Housing | mean std of shape functions: 0.9921 -> 0.1987 | Helped |
| add pairwise interactions via shared bivariate bases (NB2M vs NBM) (ablation) | NBM · CA Housing | RMSE: 0.5638 -> 0.4779 | Helped |
| add pairwise interactions (NB2M vs NBM) (ablation) | NBM · CoverType | acc@1: 0.7369 -> 0.8908 | Helped |
| add pairwise interactions (NB2M vs NBM) (ablation) | NBM · iNaturalist Birds | acc@1: 0.4227 -> 0.4684 | Helped |
| NB2M versus black-box MLP | NB2M · CoverType | acc@1: 0.9694 vs 0.8908 | Hurt |
| add triplet interactions (NB3M) (ablation) | NB2M · CoverType | acc@1: 0.8908 -> 0.9634 | Helped |
| NBM versus NODE-GAM | NODE-GAM · MIMIC-II, Credit, Click, Epsilon, Higgs, Microsoft, Yahoo, Year | AUROC/error/MSE: MIMIC-II 0.8320 -> 0.8549; Year 85.09 vs 85.10 | Helped |
| vary number of bases B (ablation) | NBM · Newsgroups, CoverType, CUB | accuracy | No clear effect |
Closest papers by meaning
- Additive Models Explained: A Computational Complexity Approach (NeurIPS 2025 · similarity 0.87)
- Gaussian Process Neural Additive Models (AAAI 2024 · similarity 0.85)
- Explainable Neural Networks with Guarantee: A Sparse Estimation Approach (AAAI 2025 · similarity 0.82)
- Interpretable Generalized Additive Models for Datasets with Missing Values (NeurIPS 2024 · similarity 0.82)
- Interpretable Next-token Prediction via the Generalized Induction Head (NeurIPS 2025 · similarity 0.82)
- QPM: Discrete Optimization for Globally Interpretable Image Classification (ICLR 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.