Research map / Architectures and efficiency / RNNs and normalization
Learning Interpretable Rules for Scalable Data Representation and Classification
Zhuo Wang, Wei Zhang, Ning Liu, Jianyong Wang
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 |
|---|---|---|---|
| add novel logical activation functions (NLAF) instead of original ones (ablation) | RRL with Gradient Grafting · activity and fashion (large) | training loss | Improved |
| add novel logical activation functions (NLAF) (ablation) | RRL with Gradient Grafting · facebook (sparse, large) | training loss | No clear effect |
| add novel logical activation functions (NLAF) (ablation) | RRL · GPU memory and time | GPU memory / time | Improved |
| use Gradient Grafting instead of STE (ablation) | RRL (NLAF) · six data sets | training loss | Improved |
| use Gradient Grafting instead of ProxQuant or Random Binarization (ablation) | RRL (NLAF) · six data sets | training loss | Improved |
| add hierarchical Gradient Grafting instead of single grafting (ablation) | deep RRL · two data sets, 3-4 logical layers | training loss | Improved |
| increase logical layers from one or two to three (hierarchical grafting) (ablation) | RRL · two data sets | training loss | Improved |
| RRL versus XGBoost | RRL · 14 datasets | average rank: 2.50 vs 3.50 | Improved |
| RRL versus CRS | RRL · facebook | F1 macro (%): 90.11 vs 11.38 | Improved |
| RRL versus C4.5 | RRL · 14 datasets | average rank: 2.50 vs 11.57 | Improved |
| RRL versus XGBoost on fashion | RRL · fashion | F1 macro (%): 89.64 vs 89.82 | Hurt |
| RRL versus transformers FT and SAINT on tic-tac-toe | RRL · tic-tac-toe | F1 macro (%): 100.0 vs 97.84 and 96.95 | Improved |
Closest papers by meaning
- Scalable Rule-Based Representation Learning for Interpretable Classification (NeurIPS 2021 · similarity 0.99)
- Learning Accurate and Interpretable Decision Rule Sets from Neural Networks (AAAI 2021 · similarity 0.86)
- Learning Interpretable Decision Rule Sets: A Submodular Optimization Approach (NeurIPS 2021 · similarity 0.85)
- Multi-value Rule Sets for Interpretable Classification with Feature-Efficient Representations (NeurIPS 2018 · similarity 0.85)
- A Bayesian Framework for Learning Rule Sets for Interpretable Classification (JMLR 2017 · similarity 0.84)
- Multistage Defer Trees for Hybrid Interpretability: If at First You Can't Succeed, Tree Again (NeurIPS 2026 · 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.
