Research map / Image generation / Autoregressive and latent generation
FR: Folded Rationalization with a Unified Encoder
Wei Liu, Haozhao Wang, Jun Wang, Ruixuan Li, Chao Yue, YuanKai Zhang
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 |
|---|---|---|---|
| Share the generator and predictor encoder | Rationalization on Beer Reviews and Hotel Reviews · Six aspects in the main comparison | F1: 72.5 -> 82.8; 63.2 -> 73.4; 65.5 -> 67.0; 53.1 -> 57.1; 43.3 -> 44.8; 33.7 -> 38.7 | Improved |
| Share the generator and predictor encoder | Rationalization at low sparsity · Beer Reviews — Appearance | F1: 64.7 -> 63.2 | Hurt |
| Share the generator and predictor encoder | Rationalization at low sparsity · Beer Reviews — Aroma and Palate | F1: 57.3 -> 68.2; 51.7 -> 61.8 | Improved |
| Share the generator and predictor encoder | Rationalization with a skewed predictor · Beer Reviews — Aroma and Palate, skew10–skew20 | F1: 61.5 -> 72.8; 49.3 -> 69.6; 11.0 -> 70.6; 5.5 -> 57.7; 1.3 -> 54.5; 0.6 -> 52.9 | Improved |
| Share the generator and predictor encoder | Rationalization with a deliberately skewed generator · Beer Reviews — Palate, skew60.0–skew75.0 | F1: 43.9 -> 59.6; 42.7 -> 63.4; 10.8 -> 58.1; 8.8 -> 51.0 | Improved |
| Share only the first GRU layer rather than all three (ablation) | Three-layer encoder rationalization · Beer-Aroma | F1: 74.9 -> 65.4 | Hurt |
| Share the first two GRU layers rather than all three (ablation) | Three-layer encoder rationalization · Beer-Aroma | F1: 74.9 -> 74.2 | Hurt |
| Share only the third GRU layer rather than all three (ablation) | Three-layer encoder rationalization · Beer-Aroma | F1: 74.9 -> 64.1 | Hurt |
| Share the second and third GRU layers rather than all three (ablation) | Three-layer encoder rationalization · Beer-Aroma | F1: 74.9 -> 71.7 | Hurt |
| Share only the first GRU layer rather than all three (ablation) | Three-layer encoder rationalization · Hotel-Cleanliness | F1: 38.3 -> 7.6 | Hurt |
| Share the first two GRU layers rather than all three (ablation) | Three-layer encoder rationalization · Hotel-Cleanliness | F1: 38.3 -> 21.8 | Hurt |
| Share only the third GRU layer rather than all three (ablation) | Three-layer encoder rationalization · Hotel-Cleanliness | F1: 38.3 -> 34.0 | Hurt |
Closest papers by meaning
- Learning Robust Rationales for Model Explainability: A Guidance-Based Approach (AAAI 2024 · similarity 0.87)
- Distilling Structured Rationale from Large Language Models to Small Language Models for Abstractive Summarization (AAAI 2025 · similarity 0.84)
- Tailoring Self-Rationalizers with Multi-Reward Distillation (ICLR 2024 · similarity 0.84)
- Fast and Fluent Diffusion Language Models via Convolutional Decoding and Rejective Fine-tuning (NeurIPS 2025 · similarity 0.83)
- DEPT: Decoupled Embeddings for Pre-training Language Models (ICLR 2025 · similarity 0.83)
- Automated Assessment of Fidelity and Interpretability: An Evaluation Framework for Large Language Models’ Explanations (Student Abstract) (AAAI 2024 · 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.
