Research map / Probabilistic and scientific modeling / Molecules and proteins
Structured Reordering for Modeling Latent Alignments in Sequence Transduction
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 |
|---|---|---|---|
| add structured reordering module (Hard-ReMoto) to SSNT (ablation) | SSNT · GeoQuery EN Temp | exact-match accuracy: 38.7 -> 43.2 | Helped |
| add structured reordering module (Hard-ReMoto) to SSNT (ablation) | SSNT · GeoQuery ZH Temp | exact-match accuracy: 23.8 -> 45.7 | Helped |
| add structured reordering module (Hard-ReMoto) to SSNT (ablation) | SSNT · GeoQuery EN IID | exact-match accuracy: 75.3 -> 75.2 | No clear effect |
| use hard Gumbel permutation instead of soft marginal permutation (ablation) | ReMoto · GeoQuery ZH Temp | exact-match accuracy: 30.3 -> 45.7 | Helped |
| use hard Gumbel permutation instead of soft marginal permutation (ablation) | ReMoto · Arithmetic Len | accuracy: 86.9 -> 83.3 | Hurt |
| use hard Gumbel permutation instead of soft marginal permutation (ablation) | ReMoto · EN-JA Len | BLEU: 27.5 -> 28.7 | Helped |
| separate parameters for alignment and decoding instead of shared input embeddings (ablation) | Soft-ReMoto · Arithmetic Len | accuracy: 40.9 -> 86.9 | Helped |
| separate parameters for alignment and decoding instead of shared input embeddings (ablation) | Soft-ReMoto · SCAN-SP Len | accuracy: 100.0 -> 100.0 | No clear effect |
| structured separable-permutation reordering instead of Gumbel-Sinkhorn token permutation (ablation) | Soft-ReMoto · Arithmetic Len | accuracy: 8.8 -> 86.9 | Helped |
| structured reordering instead of local reordering module combined with SSNT | SSNT · EN-JA Len | BLEU: 27.1 -> 28.7 | Helped |
| soft marginal reordering vs seq2seq on length split | Seq2Seq · GeoQuery EN Len | exact-match accuracy: 21.8 -> 19.8 | Hurt |
| add Hard-ReMoto to seq2seq | Seq2Seq · ZH-EN Len | BLEU: 18.1 -> 19.5 | Helped |
Closest papers by meaning
- The Missing Alignment Link of In-context Learning on Sequences (ICML 2025 · similarity 0.86)
- Leveraging Importance Sampling to Detach Alignment Modules from Large Language Models (NeurIPS 2025 · similarity 0.86)
- Long-Short Alignment for Effective Long-Context Modeling in LLMs (ICML 2025 · similarity 0.86)
- Aligner: Efficient Alignment by Learning to Correct (NeurIPS 2024 · similarity 0.85)
- Aligning Large Language Models with Representation Editing: A Control Perspective (NeurIPS 2024 · similarity 0.85)
- Flexible Realignment of Language Models (NeurIPS 2025 · similarity 0.85)
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.