Research map / Architectures and efficiency / RNNs and normalization
Decision-Guided Weighted Automata Extraction from Recurrent 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 decision-guided WFA extraction instead of WL* | RNN extraction on SPiCe · SPiCe (average) | CR: 0.59 -> 0.75 | Helped |
| Use decision-guided WFA extraction instead of WL* | RNN extraction on SPiCe · SPiCe (average) | NDCG: 0.73 -> 0.79 | Helped |
| Use decision-guided WFA extraction instead of WL* | RNN extraction on UHL · UHL (average) | CR: 0.92 -> 0.94 | Helped |
| Use decision-guided WFA extraction instead of WL* | RNN extraction on UHL · UHL (average) | NDCG: 1.00 -> 0.97 | Hurt |
| Add 2-state composition (ablation) | WFA extraction on SPiCe · SPiCe (average) | CR: 0.75 -> 0.83 | Helped |
| Add 2-state composition (ablation) | WFA extraction on SPiCe · SPiCe (average) | NDCG: 0.79 -> 0.81 | Helped |
| Add 2-state composition (ablation) | WFA extraction on UHL · UHL (average) | CR: 0.94 -> 0.94 | No clear effect |
| Add synonym transitions for unseen tokens rather than uniform transitions (ablation) | WFA extraction on QC · QC | CR: 0.50 -> 0.77 | Helped |
| Add synonym transitions for unseen tokens rather than uniform transitions (ablation) | WFA extraction on Toxic · Toxic | CR: 0.77 -> 0.86 | Helped |
| Extract states from probabilistic outputs rather than internal states (ablation) | WFA extraction on SPiCe · SPiCe (average) | CR: 0.75 -> 0.62 | Hurt |
| Extract states from probabilistic outputs rather than internal states (ablation) | WFA extraction on UHL · UHL (average) | CR: 0.94 -> 0.95 | Helped |
| Extract states from probabilistic outputs rather than internal states (ablation) | WFA extraction on QC and Toxic · QC and Toxic | CR: 0.77 -> 0.75; 0.86 -> 0.83 | Hurt |
Closest papers by meaning
- Weighted Automata Extraction from Recurrent Neural Networks via Regression on State Spaces (AAAI 2020 · similarity 0.92)
- Extracting Automata from Recurrent Neural Networks Using Queries and Counterexamples (ICML 2018 · similarity 0.90)
- State-Regularized Recurrent Neural Networks (ICML 2019 · similarity 0.90)
- Learning Useful Representations of Recurrent Neural Network Weight Matrices (ICML 2024 · similarity 0.87)
- M$^2$RNN: Non-Linear RNNs with Matrix-Valued States for Scalable Language Modeling (NeurIPS 2026 · similarity 0.85)
- Learning dynamics in linear recurrent neural networks (ICML 2025 · 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.