Research map / Learning theory and algorithms / Prediction and decision losses
Regional Tree Regularization for Interpretability in Deep Neural Networks
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 |
|---|---|---|---|
| use regional (L0/LSP) tree regularization instead of global tree regularization | MLP · toy example | test accuracy: 0.8454 -> 0.9308 | Helped |
| penalize max regional APL (L0) instead of sum of regional APLs (L1) (ablation) | regional tree regularization · toy example | test accuracy: 0.9168 -> 0.9287 | Helped |
| use SparseMax over regional APLs instead of hard max (ablation) | L0 regional tree regularization · UCI, Critical Care, HIV | training speed and accuracy/APL: about 900 epochs, 9 times longer, for L0; ten-fold fewer epochs with SparseMax | Helped |
| add Dirichlet data augmentation for surrogate training (ablation) | APL surrogate · surrogate APL prediction | mean MSE: 0.069 -> 0.015 | Helped |
| fix CART random seed (deterministic training) (ablation) | APL surrogate · surrogate APL prediction | mean MSE: 0.116 -> 0.024 | Helped |
| add reduced-error pruning of decision trees (ablation) | APL computation · toy green-vs-blue task | decision boundary quality | Helped |
| add region index as feature in global tree regularization | global tree regularization · UCI datasets | performance and simulability | No clear effect |
| use regional instead of global tree regularization | MLP · UCI datasets (Bank, Gamma, Adult, Wine) | F1 at low APL | Helped |
| use regional tree regularization | MLP · MIMIC critical care and EUResist HIV | AUC vs APL tradeoff | Helped |
Closest papers by meaning
- Beyond Sparsity: Tree Regularization of Deep Models for Interpretability (AAAI 2018 · similarity 0.91)
- Feature Learning for Interpretable, Performant Decision Trees (NeurIPS 2023 · similarity 0.87)
- ViTree: Single-Path Neural Tree for Step-Wise Interpretable Fine-Grained Visual Categorization (AAAI 2024 · similarity 0.85)
- Multistage Defer Trees for Hybrid Interpretability: If at First You Can't Succeed, Tree Again (NeurIPS 2026 · similarity 0.85)
- Near-Optimal Decision Trees in a SPLIT Second (ICML 2025 · similarity 0.84)
- Generative Model for Decision Trees (AAAI 2024 · similarity 0.84)
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.