Research map / Learning theory and algorithms / Prediction and decision losses
Scalable Interpretability via Polynomials
Abhimanyu Dubey, Filip Radenovic, Dhruv Mahajan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Use SPAM-Neural with 8 subnets instead of the original SPAM-Neural configuration (ablation) | SPAM-Neural · Cover Type (CovType) | Accuracy: 0.8857 -> 0.9405 | Improved |
| Increase interaction degree beyond 3 (ablation) | SPAM · CoverType and iNaturalist Birds | Accuracy | Improved |
| Increase cumulative rank (ablation) | SPAM · 20 Newsgroups and iNaturalist Birds | Accuracy | No clear effect |
| Use L1 sparsity on basis vectors, retaining only 6% of possible pairwise interactions (ablation) | SPAM (degree 2, rank 800) · CUB-200 | Accuracy: 6% | No clear effect |
| Use pairwise interactions rather than unary-only terms | SPAM · CUB-200 | Accuracy: 0.7451 -> 0.7786 | Improved |
| Use pairwise interactions rather than unary-only terms | SPAM · 20 Newsgroups | Accuracy: 0.8238 -> 0.8472 | Improved |
| Use pairwise interactions rather than unary-only terms | SPAM · iNaturalist Birds | Accuracy: 0.3932 -> 0.4605 | Improved |
| Use pairwise interactions rather than unary-only terms | SPAM · CO114 | mAP: 0.1917 -> 0.2361 | Improved |
| Show SPAM explanations instead of LIME explanations | CUB-200 explanation prediction task · CUB-200 | Mean user accuracy: 0.65 -> 0.71 | Improved |
| Increase explanation length beyond 7 terms (ablation) | SPAM-Linear (Order 2) and LIME explanations · CUB-200 explanation prediction task | Mean user accuracy | Hurt |
Closest papers by meaning
- Additive Models Explained: A Computational Complexity Approach (NeurIPS 2025 · similarity 0.86)
- Interpretable Generalized Additive Models for Datasets with Missing Values (NeurIPS 2024 · similarity 0.85)
- Explainable Neural Networks with Guarantee: A Sparse Estimation Approach (AAAI 2025 · similarity 0.83)
- Gaussian Process Neural Additive Models (AAAI 2024 · similarity 0.83)
- InstaSHAP: Interpretable Additive Models Explain Shapley Values Instantly (ICLR 2025 · similarity 0.83)
- TMPNN: High-Order Polynomial Regression Based on Taylor Map Factorization (AAAI 2024 · similarity 0.83)
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.
