Research map / Image generation / Fast diffusion sampling
GryphOne: Symbol-Aware Masked Diffusion for Structural Refinement in Offline Handwritten Mathematical Expression Recognition
Takaya Kawakatsu, Ryo Ishiyama
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 symbol-aware tokenization (SAT), without RMML (ablation) | GryphOne · MathWriting valid (T=50) | CER: 4.75 -> 4.68 | Improved |
| add SAT, with RMML (ablation) | GryphOne · MathWriting valid (T=50) | CER: 4.67 -> 4.63 | Improved |
| add random-masking mutual learning (RMML), without SAT (ablation) | GryphOne · MathWriting valid (T=50) | CER: 4.75 -> 4.67 | Improved |
| add RMML, with SAT (ablation) | GryphOne · MathWriting valid (T=50) | CER: 4.68 -> 4.63 | Improved |
| add SAT, without RMML (small T) (ablation) | GryphOne · MathWriting valid (T=2) | SER: 3.71 -> 2.21 | Improved |
| add RMML, without SAT (large T) (ablation) | GryphOne · MathWriting valid (T=50) | SER: 0.81 -> 0.59 | Improved |
| add SAT, without RMML (large T) (ablation) | GryphOne · MathWriting valid (T=50) | SER: 0.81 -> 0.98 | Hurt |
| increase diffusion steps from T=1 to T=50 | GryphOne · MathWriting valid | EM: 68.7 -> 71.0 | Improved |
| increase diffusion steps from T=10 to T=50 | GryphOne · MathWriting valid | FPS: 73.7 -> 21.2 | Hurt |
| replace mask-predict decoding with diffusion refinement (T=10) | GryphOne · MathWriting valid | EM: 68.2 -> 70.6 | Improved |
| replace autoregressive ICAL with GryphOne-50 | ICAL · MathWriting valid | EM: 63.4 -> 71.0 | Improved |
| replace ICAL with GryphOne-50, at most 1 error metric | ICAL · MathWriting valid | expressions with at most 1 error: 74.5 -> 85.0 | Improved |
Closest papers by meaning
- Uni-MuMER: Unified Multi-Task Fine-Tuning of Vision-Language Model for Handwritten Mathematical Expression Recognition (NeurIPS 2025 · similarity 0.90)
- SSAN: A Symbol Spatial-Aware Network for Handwritten Mathematical Expression Recognition (AAAI 2025 · similarity 0.89)
- Generating Handwritten Mathematical Expressions From Symbol Graphs: An End-to-End Pipeline (CVPR 2024 · similarity 0.88)
- TAMER: Tree-Aware Transformer for Handwritten Mathematical Expression Recognition (AAAI 2025 · similarity 0.87)
- SMILE: Bridging Continuous Optimization and Discrete Symbolic Recovery (NeurIPS 2026 · similarity 0.83)
- A Neural-Guided Dynamic Symbolic Network for Exploring Mathematical Expressions from Data (ICML 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.
