Research map / Time series / Causal and generative time series
Beyond Sparsity: Tree Regularization of Deep 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 |
|---|---|---|---|
| Train with tree regularization rather than L1 or L2 regularization | deep time-series models · Sepsis, TIMIT, HIV, and signal-and-noise HMM | AUC at low average path length: 0.05-0.1 AUC gains on Sepsis and TIMIT; 0.03 and 0.15 on HIV CD4 | Helped |
| Increase tree-regularization strength (ablation) | GRU (20) · Signal-and-noise HMM | Average Path Length: 28.1850 -> 9.1127 | Helped |
| Increase tree-regularization strength (ablation) | GRU (20) · Signal-and-noise HMM | AUC (Test): 0.94226 -> 0.90555 | Hurt |
| Apply strong tree regularization to a GRU-HMM residual (ablation) | GRU-HMM (5/15) · Signal-and-noise HMM | Average Path Length: 36.7188 -> 5.231 | Helped |
| Apply tree regularization to the GRU residual (ablation) | GRU-HMM (5/15) · Signal-and-noise HMM | AUC (Test): 0.94857 -> 0.95769 | Mixed |
| Add a tree-regularized GRU residual to an HMM | HMM · Mechanical ventilation (Sepsis) | AUC: path length 28 to reach AUC of 0.88; GRU alone requires path length 60 | Helped |
| Increase tree-regularization strength (ablation) | GRU (75) · TIMIT stop phonemes | Average Path Length: 66.2801 -> 28.4240 | Helped |
| Increase tree-regularization strength (ablation) | GRU (75) · TIMIT stop phonemes | AUC: 0.9517 -> 0.8998 | Hurt |
| Use elastic net instead of L1 or L2 (ablation) | GRU · time-series tasks | AUC versus average path length | No clear effect |
| Use a 100-hidden-unit surrogate rather than the 25-unit surrogate (ablation) | tree-regularization surrogate · surrogate path-length estimation | surrogate accuracy | Hurt |
| Augment surrogate training with randomly sampled weight vectors and retrain it periodically (ablation) | tree-regularization surrogate · surrogate path-length estimation | tracking accuracy | Helped |
| Use an average-path-length tree penalty rather than L1 or L2 penalties | MLP classifier · 2D Parabola | AUC at path lengths between 0 and 5 | Helped |
Closest papers by meaning
- Regional Tree Regularization for Interpretability in Deep Neural Networks (AAAI 2020 · similarity 0.91)
- Feature Learning for Interpretable, Performant Decision Trees (NeurIPS 2023 · similarity 0.86)
- Regularization Learning Networks: Deep Learning for Tabular Datasets (NeurIPS 2018 · similarity 0.86)
- Multistage Defer Trees for Hybrid Interpretability: If at First You Can't Succeed, Tree Again (NeurIPS 2026 · similarity 0.85)
- A Textbook Remedy for Domain Shifts: Knowledge Priors for Medical Image Analysis (NeurIPS 2024 · similarity 0.83)
- AutoElicit: Using Large Language Models for Expert Prior Elicitation in Predictive Modelling (ICML 2025 · 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.