Research map / Architectures and efficiency / RNNs and normalization
Understanding Interlocking Dynamics of Cooperative Rationalization
Mo Yu, Yang Zhang, Shiyu Chang, Tommi S. Jaakkola
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 |
|---|---|---|---|
| replace RNP with A2R (add attention-based predictor and JS gap term) | RNP · BeerAdvocate Palate | rationale F1: 51.0 -> 58.0 | Improved |
| replace RNP with A2R | RNP · BeerAdvocate Appearance | rationale F1: 72.8 -> 72.9 | No clear effect |
| replace RNP with A2R | RNP · MovieReview | rationale F1: 24.1 -> 34.9 | Improved |
| replace RNP with A2R under strong skew pre-training | RNP · Beer-Skew Aroma Skew20 | rationale F1: 11.0 -> 46.3 | Improved |
| replace RNP with A2R under strong skew pre-training | RNP · Beer-Skew Palate Skew20 | rationale F1: 0.6 -> 0.6 | No clear effect |
| replace RNP with A2R under spurious punctuation bias | RNP · Beer-Biased Palate 0.75 | rationale F1: 0.6 -> 47.9 | Improved |
| set lambda to 0 (remove JS gap term, soft attention only) (ablation) | A2R · Beer-Skew Palate Skew20 | rationale F1: 2.2% F1 with 97.3% X1 selection | No clear effect |
| vary lambda from 1e-3 to 10 (ablation) | A2R · Beer-Skew/Biased | rationale F1 | No clear effect |
| anneal lambda from 0 to 1.0 over 10 epochs (ablation) | A2R · Beer synthetic settings | rationale F1 | Hurt |
Closest papers by meaning
- Adversarial Cooperative Rationalization: The Risk of Spurious Correlations in Even Clean Datasets (ICML 2025 · similarity 0.84)
- Towards Faithful Explanations: Boosting Rationalization with Shortcuts Discovery (ICLR 2024 · similarity 0.84)
- From Lazy to Rich: Exact Learning Dynamics in Deep Linear Networks (ICLR 2025 · similarity 0.84)
- In-Context Learning Strategies Emerge Rationally (NeurIPS 2025 · similarity 0.83)
- In-Context Linear Regression Demystified: Training Dynamics and Mechanistic Interpretability of Multi-Head Softmax Attention (ICML 2025 · similarity 0.83)
- Alternating Gradient Flows: A Theory of Feature Learning in Two-layer Neural Networks (NeurIPS 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.
