Research map / Image generation / Fast diffusion sampling
Insertion Based Sequence Generation with Learnable Order Dynamics
Dhruvesh Patel, Benjamin Rozonoyer, Gaurav Pandey, Tahira Naseem, Ramón Fernandez Astudillo, Andrew McCallum
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 |
|---|---|---|---|
| add learnable order dynamics (LoFlexMDM instead of FlexMDM, best decoding) | FlexMDM · De novo molecules | quality (%): 62.0 -> 79.5 | Improved |
| add learnable order dynamics | FlexMDM · De novo molecules | diversity: 0.890 -> 0.850 | Hurt |
| add learnable order dynamics | FlexMDM · De novo molecules | validity (%): 98.9 -> 99.7 | Improved |
| add learnable order dynamics | FlexMDM · Fragment-constrained: motif extension | quality (%): 46.9 -> 53.6 | Improved |
| add learnable order dynamics | FlexMDM · Fragment-constrained: linker design | quality (%): 50.8 -> 51.7 | Improved |
| add learnable order dynamics | FlexMDM · Fragment-constrained: scaffold decoration | quality (%): 39.0 -> 40.5 | Improved |
| add learnable order dynamics | FlexMDM · Fragment-constrained: superstructure generation | quality (%): 35.8 -> 37.0 | Improved |
| use confidence-based position selection for unmasking (ablation) | FlexMDM · De novo molecules | quality (%): 62.0 -> 5.5 | Hurt |
| use confidence-based position selection for unmasking (ablation) | LoFlexMDM medium aux, p=0.5 · De novo molecules | quality (%): 79.5 -> 69.3 | Hurt |
| lower nucleus cutoff p from 1.0 to 0.5 (ablation) | LoFlexMDM medium aux, no conf. · De novo molecules | quality (%): 55.1 -> 79.5 | Improved |
| increase auxiliary rate network size from xtiny to medium (ablation) | LoFlexMDM, no conf., p=0.5 · De novo molecules | quality (%): 79.4 -> 79.5 | No clear effect |
| fix unmasking rate b_unmask instead of trainable (ablation) | LoFlexMDM · Star graph hard | exact match accuracy | Improved |
Closest papers by meaning
- DMol: A Highly Efficient and Chemical Motif-Preserving Molecule Generation Platform (NeurIPS 2025 · similarity 0.86)
- Unlocking Guidance for Discrete State-Space Diffusion and Flow Models (ICLR 2025 · similarity 0.86)
- Steering Masked Discrete Diffusion Models via Discrete Denoising Posterior Prediction (ICLR 2025 · similarity 0.86)
- Graph Diffusion that can Insert and Delete (NeurIPS 2025 · similarity 0.85)
- Train for the Worst, Plan for the Best: Understanding Token Ordering in Masked Diffusions (ICML 2025 · similarity 0.85)
- Text-Guided Molecule Generation with Diffusion Language Model (AAAI 2024 · 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.
