Research map / Vision-language and multimodal / Cross-modal learning
MM-Spectrum: Multimodal Multi-spectral Molecular Structural Elucidation with a Stable MoE Framework
Hai-tao Yu, Min Nan, Zheng Fang, Hongyu Zhan, Yusen Tan, Yuhan Wang, Jun Xia
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 Router-aware signals (ablation) | MM-Spectrum · Spectra benchmark (NMR+MS+IR) | Top-1 accuracy: 72.57 -> 76.04 | Improved |
| add Tag routing (ablation) | MM-Spectrum · Spectra benchmark (NMR+MS+IR) | Top-1 accuracy: 71.23 -> 76.04 | Improved |
| add both Router-aware signals and Tag routing (ablation) | MM-Spectrum · Spectra benchmark (NMR+MS+IR) | Top-1 accuracy: 70.11 -> 76.04 | Improved |
| add Dynamic activation (ablation) | MM-Spectrum · Spectra benchmark (NMR+MS+IR) | Top-1 accuracy: 74.41 -> 76.04 | Improved |
| add Interaction experts (ablation) | MM-Spectrum · Spectra benchmark (NMR+MS+IR) | Top-1 accuracy: 74.39 -> 76.04 | Improved |
| add Shared experts (ablation) | MM-Spectrum · Spectra benchmark (NMR+MS+IR) | Top-1 accuracy: 75.01 -> 76.04 | Improved |
| add computation-aware regularizer (ablation) | MM-Spectrum · Spectra benchmark (NMR+MS+IR) | Top-1 accuracy: 74.81 -> 76.04 | Improved |
| add heterogeneous expert capacities (ablation) | MM-Spectrum · Spectra benchmark (NMR+MS+IR) | Top-1 accuracy: 74.07 -> 76.04 | Improved |
| add heterogeneous capacities without regularizer vs neither (ablation) | MM-Spectrum · Spectra benchmark (NMR+MS+IR) | Top-1 accuracy: 74.24 vs 74.07 | No clear effect |
| replace dense concatenation with modality-aware sparse MoE | Dense concatenation · Spectra benchmark (NMR+MS+IR) | Top-1 accuracy: 44.29 -> 76.04 | Improved |
| add IR and MS to NMR with dense concatenation | Dense concatenation · Spectra benchmark (NMR+MS+IR) | Top-1 accuracy: 69.83 -> 44.29 | Hurt |
| use Gradient Blending instead of plain dense concatenation | Dense concatenation · Spectra benchmark (NMR+MS+IR) | Top-1 accuracy: 44.29 -> 46.78 | Improved |
Closest papers by meaning
- Unraveling Molecular Structure: A Multimodal Spectroscopic Dataset for Chemistry (NeurIPS 2024 · similarity 0.87)
- MS-BART: Unified Modeling of Mass Spectra and Molecules for Structure Elucidation (NeurIPS 2025 · similarity 0.87)
- GLACIER: Rethinking Mass Spectrum Prediction as an Object Detection Problem (NeurIPS 2026 · similarity 0.84)
- Can LLMs Solve Molecule Puzzles? A Multimodal Benchmark for Molecular Structure Elucidation (NeurIPS 2024 · similarity 0.84)
- MassSpecGym: A benchmark for the discovery and identification of molecules (NeurIPS 2024 · similarity 0.83)
- Atomic Diffusion Models for Small Molecule Structure Elucidation from NMR Spectra (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.
